Polynomial Automorphisms and Related Top PDF

Arno van den Essen
Ha Huy Vui
Hanspeter Kraft
Peter Russell
David Wright
Polynomial Automorphisms
and Related Topics
Lecture Notes of the International School and Workshop
ICPA2006, October 2006, Institute of Mathematics
Hanoi, Vietnam
Editors:
Hymann Bass
Nguyen Van Chau
Stefan Maubach
PUBLISHING HOUSE FOR SCIENCE AND TECHNOLOGY

Publishing House for Science and Technology
18 Hoang Quoc Viet Road, Cau Giay, Hanoi, Vietnam
Tel: 84-4-2149041; Fax: 84-4-7564483
Email: nxb@vap.ac.vn; http:// www.vap.ac.vn
Authors:
Arno van den Essen
Ha Huy Vui
Hanspeter Kraft
Peter Russell
David Wright
Polynomial Automorphisms and Related Topics

Lecture Notes of the International School and Workshop
ICPA2006, October 2006, Institute of Mathematics, Hanoi, Vietnam
Editors:
Hyman Bass
Nguyen Van Chau
Stefan Maubach
2000 Mathematics Subject Classification: 14E xx, 14 R xx.
Responsibility for Publication: Nguyen Khoa Son

Technical Edited: Tran Phuong Dong
Printed in Printed House of Science and Technology
Registered Ref. No: 776-2007/CXB/008-04/KHTNCN
Dated: September 24, 2007.
Contents
Preface III
Suggested Notation for Groups of Polynomial Automorphisms IX
Chapter 1. Polynomial automorphism groups 1
1. What is a polynomial automorphism? 1
2. Recognizing an automorphism 2
3. The case n = 1, R a domain 2
4. Notation 3
5. Canonical homomorphisms 3
6. Subgroups 4
7. Formal inverse 5
8. More subgroups 6
9. More on n = 1 6
10. Elementary and tame automorphisms 7
11. Stable tameness 9
12. The case n = 2, R a field 10
13. Structure of GA2 (k), k a field 16
14. Structure of T2(R), R a domain 18
15. The case n = 3 19
Chapter 2. Perhaps the Jacobian Conjecture is simple 21
Introduction 21
1. The classical formulation 21
2. The Rolle formulation 22
3. The degree of the inverse 26
4. Zhao’s vanishing conjecture 30
5. The Dixmier Conjecture 33
6. Endomorphisms of canonical Poisson algebras over commutative
rings 40
Chapter 3. Basics from algebraic geometry 43
1. Affine varieties 43
I
II CONTENTS
2. Morphisms 60
3. Dimension 69
4. Tangent Spaces and Differentials 81
5. Normal Varieties and Divisors 101
Chapter 4. Embedding problems in affine algebraic geometry 113
1. The problem and first examples 113
2. The case of affine spaces 114
3. The high codimension case 115
4. General and generic fibers 117
5. More on n=3 121
6. Curves with one place at infinity 123
Appendix chapter 4 part 1: Notation 127
Appendix chapter 4 part 2: More on general and generic fibers 129
Appendix chapter 4 part 3: Plane algebroid curves 131
Chapter 5. Bifurcation set of the global Milnor fibration 137
1. Introduction and definitions 137
2. Case n = 2 138
3. Rabier-Jelonek Fibration Theorem 144
4. Jelonek set 148
Bibliography 153
List of notations 157
Index 159
Preface
The International School and Workshop “Polynomial Automorphisms and

Related Topics” was held at the Institute of Mathematics, Hanoi, Vietnam,
on October 9-20, 2006. Central to the theme of this International School and
Workshop were some venerable, but still largely intractable, (linearization)
problems of affine algebraic geometry: the Jacobian Conjecture (lineariz-
ability of the criterion for invertibility); generators of the automorphism
group of affine space (what more than linear and triangular automorphisms
is needed?); algebraic group actions on affine space (linearizability ques-
tions); linearizability of embeddings of one affine space in another; etc. An
extraordinary variety of techniques and approaches have been developed for
treating these questions. The workshop assembled many of the world’s ex-
perts to present the state of the art of research in this interesting and very
active area.
The preceding International School enlisted some of the best expositors
in the field to provide an excellent and systematic foundation for graduate
students and young researchers to gain entry. It is the notes of those lecture
courses that are assembled here. They should provide a valuable compre-
hensive and accessible reference work for the field for some time to come. We
have the organizers, the Institute of Mathematics, Vietnam, and the Abdus
Salam ICTP to thank for making this very successful event and publication
possible.
It is noteworthy, and highly appropriate, that this event was staged in
Vietnam. First of all, it paid tribute to the vibrant research environment
being created in Vietnam under the leadership and vision of Vietnamese
mathematicians like Le Dung Trang, Ha Huy Vui and Nguyen Van Chau. It
is hoped that our meeting in Hanoi provided young researchers in Vietnam
and the other developing countries with some of the much deserved resources,
encouragement, and international connection that are not so easily available
at this stage of development in their country.
It must be said that all of the visiting participants profited as well, not
only from the intellectual stimulation of meeting our Vietnamese mathemat-
ical colleagues, but also from the warm hospitality of our hosts, and from
the wonderful food, culture, and environment of their beautiful country.
Ann Arbor, October, 2007

Hyman Bass
III
IV
International School and Workshop

Polynomial Automorphisms and Related Topics
October 9-20, 2006, Hanoi, Vietnam
Organizations
Abdus Salam International Centre for Theoretical Physics (ICTP),

Trieste, Italy.
Institute of Mathematics, Vietnamese Academy of Science
and Technology, Hanoi, Vietnam.
Organizers
Arno Van den Essen, Radboud University of Nijmegen, Netherlands.

David Wright, Washington University in St. Louis, USA.
Le Tuan Hoa, Institute of Mathematics, Hanoi, Vietnam.
Nguyen Van Chau, Institute of Mathematics, Hanoi, Vietnam.
Scientific Committee
Hyman Bass, Michigan University, USA.

Shreeram Abhyankar, Purdue University, USA.
Arno Van den Essen, Radboud University of Nijmegen, Netherlands.
Le Dung Trang, Abdus Salam ICTP, Italy.
Cesar Camacho, IMPA, Brazil.
David Wright, Washington University in St. Louis, USA.
Ha Huy Vui, Institute of Mathematics, Hanoi, Vietnam.
Ngo Viet Trung, Institute of Mathematics, Hanoi, Vietnam.
Ha Huy Khoai, Institute of Mathematics, Hanoi, Vietnam.
Nguyen Van Chau, Institute of Mathematics, Hanoi, Vietnam.
List of mini-courses
Arno Van den Essen, Perhaps the Jacobian Conjecture is simple.

Ha Huy Vui, Bifurcation set of the global Milnor fibration.
V
Hanspeter Kraft, An introduction to algebraic transformation.

Peter Russell, Embedding problems in affine algebraic geometry.
David Wright, Polynomial automorphism groups.
List of talks1
Shreeram Abhyankar, Some thoughts on the Jacobian conjecture.

Teruo Asanuma, A1-forms of characteristic p > 0.
Hyman Bass, Asanuma Algebra.
Michiel de Bondt, A Druzkowski counter example to the linear dependence.
Phillipe Bonnet, On algebraic automorphisms and their rational invariants.
Pierrette Cassou-Nogues, About the images of a linear pencil of lines by an
affine endomorphism of C2 .
Eric Edo, Limit of family automorphisms.
Arno Van den Essen, A short proof of the equivalence of the Dixmier, Jaco-
bian and Poisson conjectures.
David Finston, Additive group actions with finitely generated invariants.
Jean-Philippe Furter, Jet groups and fat points embeddings.
Ha Huy Vui, An estimation of the number of bifurcation values for real
polynomials.
He Tong, About the matrix of power linear Keller maps.
Tasuji Kambayashi, Forms of Laurent polynomial rings and algebraic tori at
dimensions 1 and 2.
Mariusz Koras, Contractible surfaces with quotient singularities.
Hanspeter Kraft, A Galois correspondence for algebraic groups.
Tadeusz Krasinski, On Lojasiewicz exponent at infinity of polynomial
mappings.
Shigeru Kuroda, Hilbert’s fourteenth problem and algebraic extensions.
Sooraj Kuttykrisman, Stable tameness in AutK[Z] K[Z][X, Y ].
Le Dung Trang, Some thoughts about plane Jacobian Problem.
1
See Acta Math. Vietnam. 32 (2007)- Special issue for the Proceedings of the Interna-
tional School and Workshop “Polynomial Automorphism and Related Topics”, Oct. 2006,
Hanoi, Vietnam.
VI
Le Tuan Hoa, Finiteness of Hilbert functions and bounds for Castelnuovo-

Mumford regularity of initial ideals .
Kayo Masuda, Locally nilpotent derivations with slices.
Stefan Maubach, A three dimensional UFD cancellation counterexample
Masayoshi Miyanishi, Pure subalgebras of a polynomial ring and related
topics.
Walter Neumann, Splice Singularities.
Ngo Viet Trung, Bernstein’s theorem and mixed multiplicities of ideals.
Nguyen Van Chau, Singularity and non-proper value curve of polynomial
map of C2 .
Mutsuo Oka, Jacobian conjecture of two variables.
Ronen Peretz, The Jacobian algebraic set.
Pierre-Marie Poloni, Embeddings of a family of Danielewski hypersurfaces.
Peter Russell, C∗ in C2 .
Ta Le Loi, Transversality theorem in o-minimal structures.
Stephan Venereau, New bad lines and optimization of the epimorphism
theorem.
David Wright, The Jacobian conjecture as a problem in combinatorics.
Jie-Tai Yu, Automorphisms and coordinates of polynomial and free

associative algebras.
Wenhua Zhao, Noncommutative symmetric functions.

VII
The International School “Polynomial Automorphisms and Related Topics”

October 9-14, Hanoi, Vietnam
The International Workshop “Polynomial Automorphisms and Related Topics”

October 16-20, Hanoi, Vietnam
Suggested Notation for Groups of Polynomial
Automorphisms
David Wright
The notation R[n] for the polynomial ring in n variables over a commuta-
tive ring R has now become standard. As the group of R-algebra automor-
phisms of R[n] and certain of its subgroups have become the focus of much
attention, some unifying notation is greatly needed.
We often see notations such as AutR (R[X1 , . . . , Xn ]) or AutR (R[n] ) for the
full group of automorphisms, but this is akin to writing AutR (Re1 ⊕· · ·⊕Ren )
or AutR(Rn ) for the group of R-module automorphisms of the free R-module
Re1 ⊕ · · · ⊕ Ren = Rn . Instead, linear algebra and algebraic K-theory call it
the general linear group and use the simpler functorial notation GLn (R) or
GL(n, R), which presumes that a basis {e1, . . . , en } has been chosen.
A polynomial automorphism is given by a vector of polynomials F =
(F1, . . . , Fn ), with X = (X1 , . . . , Xn ) being the identity. We first observe that
in order to have the more natural group law
F ◦ G = (F1(G1 , . . . , Gn ), . . . , Fn (G1 , . . . , Gn ))
for polynomial automorphisms F = (F1, . . . , Fn ) and G = (G1 , . . . , Gn ), we

must view F and G as automorphisms of AnR = Spec(R[n] ), whose auto-
morphism group over Spec(R) is anti-isomorphic to AutR (R[n] ). Hence we
consider a polynomial automorphism to be an element of the former group.
The general automorphism group. In the spirit of linear algebra, we pro-

pose the symbol GAn (R) for the general automorphism group AutSpec(R) (AnR )
of algebraic (i.e., polynomial) automorphisms of AnR over R. This, in similar
fashion to its linear analogue, presumes that a set of variables {X1 , . . . , Xn }
has been designated. We sometime simply write GAn when R is understood
in context, and we similarly abbreviate the notations for the subgroups de-
fined below.
IX
X SUGGESTED NOTATION FOR GROUPS OF POLYNOMIAL AUTOMORPHISMS
The general linear group. We then have the general linear group GLn (R)
naturally contained in GAn (R) as the subgroup of all F = (F1, . . . , Fn ) for
which each Fi is a linear form. Here an invertible matrix A is identified with
the automorphism F given by the vector (A · X)t , where X = (X1 , . . . , Xn )t.
We can view GLn (R) anti-isomorphically as the stabilizer in R[n] of the R-
module RX1 ⊕ · · · ⊕ RXn .
The origin preserving group. The group of “origin preserving” automor-

phisms, i.e., those F ∈ GAn (R) for which F (0) = 0, is denoted by GA0n (R).
Note that GLn (R) ⊆ GA0n (R).
The tangent preserving groups. The group of “tangent preserving” au-

tomorphisms, denoted by GA1n (R), consists of all F = (F1, . . . , Fn ) for which
Fi has the form Xi + terms of order ≥ 2. More generally, for k an integer ≥ 1
we have the subgroups GAkn (R) whose F have the property Fi = Xi + terms
of order ≥ k + 1. We note that
∩k GAkn (R) = {(X1 , . . . , Xn )}.
The affine group. The affine group consists of those automorphisms of the
form F = (L1 + c1 , . . . , Ln + cn ) where L1 , . . . , Ln are linear forms in R[n] and
c1 , . . . , cn ∈ R. We denote this group by Afn (R). Note that GLn (R) ⊆ Afn (R),
and, in fact, GLn (R) = Afn (R) ∩ GA0n (R).
The group of translations. Also contained in the affine group is the group
of translations, i.e., those automorphisms of the form
F = (X1 + c1 , . . . , Xn + cn )
with c1 , . . . , cn ∈ R. We denote this subgroup by Trn (R). It is isomorphic to
the additive group Rn .
The elementary group. An automorphism is called elementary or de Jon-

quiere, if it has the form
F = (X1 , . . . , Xi−1 , Xi + f, Xi+1 , . . . , Xn ) ,
where f ∈ R[X1 , . . . , Xi−1 , Xi+1 , . . . , Xn ], for some i ∈ {1, . . . , n}. Since this
generalizes the concept of an elementary matrix, and since the group gen-
erated by elementary matrices is commonly denoted by En (R), we will de-
note by EAn (R) the group generated by all elementary automorphisms. Then
EAn (R) ⊇ En (R), and it is not hard to show that En (R) = EAn (R)∩GLn (R).
SUGGESTED NOTATION FOR GROUPS OF POLYNOMIAL AUTOMORPHISMS XI
The tame group. The group of tame automorphisms, denoted by Tn (R),

is the subgroup of GAn (R) generated by EAn (R) and GLn (R), i.e., Tn (R) =
h EAn (R), GLn (R) i.
The triangular group. Unlike the subgroups defined above, the trian-
gular subgroup depends on the order of the variables X1 , . . . , Xn , as does
the linear triangular group B2(R), sometimes called the Borel group. We
will take it to be the group of lower triangular matrices. An automorphism
F = (F1 , . . . , Fn ) is called triangular if, for i = 1, . . . , n, Fi has the form
ui Xi + f (X1 , . . . , Xi−1 ), with ui a unit in R. Such automorphisms form
a group which can be viewed as the non-linear extension of Bn (R). We
will call it the nonlinear Borel group, and denote it by BAn (R). Note that
BAn (R) ∩ GLn (R) = Bn (R).
The affine triangular group. The affine triangular group, affine Borel
group, Bfn (R) is defined as the intersection BAn (R) ∩ Afn (R). It consists of
those affine automorphisms whose linear homogeneous part is upper trian-
gular. The Jung-van der Kulk Theorem can be stated as
GA2 (k) = Af2 (k) ∗Bf2 (k) BA2 (k)
for k a field.
Fixator and stabilizer groups. For s = 1, . . . , n, the group GAn−s (R[s] ) is

realized anti-isomorphically in GAn (R) as the fixator of the set {X1 , . . . , Xs }.
This consists of all automorphisms having the form
F = (F1 , . . . , Xs , Fs+1 , . . . , Fn ).
We also have the stabilizer (again, anti-isomorphically) of the R-module
RX1 ⊕· · · ⊕ RXs consisting of F = (F1, . . . , Fn ) for which F1 , . . . , Fs have de-
gree 1 (not necessarily homogeneous). We denote this subgroup by Ws,n (R).
Observe that GAn−s (R[s] ) ⊆ Ws,n (R), and that Wn,n (R) = Afn (R), and that
W1,2(R) = BA2(R).
The stable automorphism group. Finally, we have natural “stability”

containment GAn (R) ,→ GAn+m (R) identifying F = (F1, . . . , Fn) with
F [m] = (F1, . . . , Fn, Xn+1 , . . . , Xn+m ). We define GA∞ (R) as the direct limit,
or ascending union, of
GA1 (R) ,→ GA2(R) ,→ GA3(R) ,→ · · ·
XII SUGGESTED NOTATION FOR GROUPS OF POLYNOMIAL AUTOMORPHISMS
We note that all the other groups defined here stabilize and we similarly
denote their direct limits, e.g., by T∞ (R), EA∞ (R), Ws,∞ (R).
CHAPTER 1
Polynomial automorphism groups
David Wright
1. What is a polynomial automorphism?

Let R be a commutative ring. We will write R∗ for the group of units in
R and R× for the monoid of non-zero-divisors in R.
Let n ∈ Z+ . A polynomial automorphism (over R) can be viewed as
either:
(1) an R-algebra automorphism of the polynomial ring R[X] =
R[X1 , . . . , Xn ], or contravariantly,
(2) an automorphism of the affine scheme AnR = Spec R[X] over Spec R
ϕ=ϕF
R[X1 , . . . , Xn ] −−−→ R[X1 , . . . , Xn ]
F =Fϕ
AnR ←−−− AnR
Fϕ is given by F = (F1, . . . , Fn ) ∈ R[X]n where Fi = ϕ(Xi ).
Contravariance means:
Fψ◦ϕ = Fϕ ◦ Fψ and ϕG◦F = ϕF ◦ ϕG .
We will adopt 2 as our definition, so that composition is given by
F ◦ G = (F1(G1 , . . . , Gn ), . . . , Fn (G1 , . . . , Gn ))
and the familiar Chain Rule holds:
J (F ◦ G) = J (F )|X=G · J (G)
where  
D1 F1 · · · Dn F1
J (F ) =  ... .. ..  , D = ∂ .
. . i
∂Xi
D1 Fn · · · Dn Fn
The group of automorphisms (over R in dimension n) is taken to be the
set of such maps which are invertible, under this law of composition.
1
2 1. POLYNOMIAL AUTOMORPHISM GROUPS
2. Recognizing an automorphism
How does one tell if F = (F1, . . . , Fn) ∈ R[X]n determines an automor-
phism?
If so then F −1 ◦ F = (X1 , . . . , Xn ), so
J (F −1 ◦ F ) = J (F −1)|X=F · J (F ) = In ,
hence a necessary condition is that J (F ) ∈ GLn (R[X]), i.e., det J (F ) ∈
R[X]∗ .
Sufficient? For R a field of characteristic zero (or just a Q-algebra), this
is the famous Jacobian Conjecture.
Note: If F is an automorphism, then J F |X=0 ∈ GLn (R),
JF |X=0 = (aij ),
where ai1X1 + · · · + ain Xn is the linear homogeneous summand of Fi .
So a necessary condition for F to be an automorphism is that “the linear
part of F is good”.
Proposition 1.1. F = (F1, . . . , Fn ) is an automorphism ⇐⇒
R[X1 , . . . , Xn ] = R[F1, . . . , Fn ].
Proof. =⇒ : Clear, since for G = F −1 we have X1 = Gi (F1 , . . . , Fn ).
X 7→Fi
⇐=: ϕF : R[X1 , . . . , Xn ] −−i−−→ R[X1 , . . . , Xn ] is surjective, i.e., F has a
left inverse. Show F injective. If not, choose H ∈ Ker ϕf and replace R by
the Noetherian ring R0 generated by the coefficients of F , G, and H. The
proposition now follows from the proposition below.
ϕ
Proposition 1.2. A surjective ring endomorphism A → A, where A is
Noetherian, is injective.
Proof. If not, we have
0 $ Ker ϕ $ Ker ϕ2 $ Ker ϕ3 $ · · ·
violating the Noetherian property.
3. The case n = 1, R a domain

In this case
deg (G ◦ F ) = deg G · deg F.
So if G ◦ F = X then deg G = deg F = 1 : F = uX + a, G = vX + b, and
X = G ◦ F = v(uX + a) + b = vuX + va + b
5. CANONICAL HOMOMORPHISMS 3
so u, v ∈ R∗ , v = u−1 , b = −va. Any F = uX + a with u ∈ R∗ is invertible.

u a
F ←→
0 1
gives
an isomorphism
between the polynomial automorphism group, n = 1,
∗ ∗
with ⊆ GL2(R).
0 1
4. Notation
We will sometimes write R[n] for the polynomial ring R[X1 , . . . , Xn ].
For G a group we write h · · · i for the subgroup of G generated by the
elements of G or subsets of G listed, and we write moniod [ · · · ] for the sub-
monoid so generated.
GAn (R) = group of polynomial automorphisms in dimension n over R,
i.e., automorphisms of AnR = Spec R[X1 , . . . , Xn ] over Spec R.
It is the group of invertible elements in MAn (R) = monoid of all algebraic
endomorphims of AnR over Spec R.
For F = (F1, . . . , Fn ) ∈ GAn (R) (or even MAn (R)) the degree of F ,
denoted by deg F , is the maximum of the total degrees of Fi in the variables
X1 , . . . , X n .
5. Canonical homomorphisms
(1) Functoriality (base change): A ring homomorphism α : R → R0 extends
to α̂ : R[X] → R0 [X], sending Xi to Xi , inducing a group homomorphism
GAn (R) → GAn (R0 )
F = (F1, . . . , Fn ) 7−→ (α(F1 ), . . . , α(Fn )).
(2) Stability: We have GAn (R) ,→ GAn+m (R) identifying F = (F1 , . . . , Fn )
with
F [m] = (F1, . . . , Fn , Xn+1 , . . . , Xn+m ).
We define GA∞ as the direct limit (union) of
GA1 (R) ,→ GA2(R) ,→ GA3(R) ,→ · · ·
(3) Restriction of scalars: We have GAm (R[Y1 , . . . , Yr ]) ,→ GAr+m (R) defined
by
(F1, . . . , Fm ) 7−→ (Y1 , . . . , Yr , F1, . . . , Fm).
6. Subgroups
• General linear group: GLn ,→ GAn as the stabilizer of RX1 ⊕· · ·⊕ RXn .
     t
a11 · · · a1n a11 · · · a1n X1
 ... ..
.
..  7−→  ..
. .
..
.
..   .. 
. .
an1 · · · ann an1 · · · ann Xn
• Translations: Rn ,→ GAn defined by
(b1 , . . . , bn ) 7−→ (X1 + b1, · · · , Xn + bn )
k k
b X +b
defines a homomorphism whose image is denoted by Trn (R), the subgroup
of translations.
• Affine group: It is defined as
Afn (R) = {F ∈ GAn (R) | deg F = 1}.
It consists of all F = (F1, . . . , Fn ) where
Fi = ai1 X1 + · · · + ain Xn + bi
with ai1, · · · , ain , bi ∈ R.
Moreover,
 
 a11 · · · a1n b1 

 .. .. .. .. 

Afn (R) ∼  . . . . 
= 
 a · · · a b  ⊆ GLn+1 (R)

 n1 nn n 

0 ··· 0 1
and
Afn = GLn nTrn
n
where GLn acts on Trn = R in the usual way.
• Origin preserving automorphisms:
GAon (R) = {F ∈ GAn (R) | F (0) = 0}.
This is (anti-isomorphically) the stabilizer of the ideal
(X1 , . . . , Xn )R[X1 , . . . , Xn ].
We have GLn ⊆ GAon with retraction
F 7−→ J F | X=0
7. FORMAL INVERSE 5
which is a homomorphism since

F · G 7−→ J (F · G)|X=0
= [J (F )|X=G · J (G)] |X=0
=JF |X=G(0) · JG|X=0
=JF |X=0 · J G|X=0.
Note that if F ∈ GAn and b = (F1 (0), . . . , Fn (0)), then (X − b) ◦ F =

F − b ∈ GAon , which shows
GAn = Trn · GAon = GAon · Trn
However, this is not a semidirect product. In fact, neither subgroup is normal.

In many arguments we may replace F by F − F (0) to assume F ∈ GAon ,
but one must be careful. (We will see why later.)
• Tangent preserving automorphisms: The kernel of this retraction is
denoted by GA0n (R). It consists of all F ∈ GAn of the form F = X +
higher degree terms. We have
GAon = GLn nGA0n
It follows that GAn = Afn · GA0n (not semidirect).
7. Formal inverse
Suppose P ∈ MAn (R) with “good” linear part, i.e., J P |X=0 ∈ GLn (R).
Then, there exists A ∈ Afn (R) such that F = A ◦ P = X + higher terms .
P will be an automorphism if and only if F is.
Such an F determines an automorphism of R[[X1 , . . . , Xn ]], so there exists
a unique system of power series G = (G1 , . . . , Gn ) of the form X + higher
such that G ◦ F = F ◦ G = X. G is called the formal inverse of F .
Proposition 1.3. F ∈ GAn ⇐⇒ G is a polynomial map.

Consequence:
Overring Principle. Suppose R is a subring of R0 , F ∈ MAn (R) with

good linear part over R. If F ∈ GAn (R0 ), then F ∈ GAn (R).
Proof. Consider the formal inverse.

8. More subgroups
• Fixator of {X1 , . . . , Xs }: For 1 ≤ s ≤ n, the restriction of scalars map
gives
GAn−s (R[s] ) ,→ GAn (R)
identified as the subgroup of all F of the form
F = (X1 , . . . , Xs , Fs+1 , . . . , Fn).
Example: (X, Y +X 2 ) ∈ GA2 (R) can be viewed as an element of GA1 (R[X]).
Aside: Note that GAon (R) ∩ GAn−s (R[s] ) is not contained in
GAon−s (R[s] ). (see above)
This restricts to GLn−s (R[s] ) ⊆ GAon (R).
Example: The “Cohn matrix”

1 + XY Y2
∈ GL2(R[X, Y ])
−X 2 1 − XY
gets identified with (X, Y, (1 + XY )Z + Y 2 W, −X 2Z + (1 − XY )W ) ∈
GAo4 (R).
• Stabilizer of R ⊕ RX1 ⊕ · · · ⊕ RXi : For i = 1, . . . , n, let Wi,n (R) be this
stabilizer. The containment Afi (R) ⊆ Wi,n (R) via the stability homomor-
phism (F 7→ F [n−i] ) admits the retraction Wi,n → Afi which is the restriction
from R[n] to R[i] , whose kernel is GAn−i (R[i] ). Therefore
Wi,n (R) = Afi (R) n GAn−i (R[i] ) .

We have:
Wn,n (R) = Afn (R)
and for R reduced:
Wn−1,n (R) = Afn−1 (R) n Af1(R[n−1] ).
9. More on n = 1
Suppose R is a reduced ring, F = (a0 +a1 X+· · ·+ad X d ) ∈ GA1 (R), p ⊆ R
a prime ideal. Base change gives F = (a0 + a1 X + · · · + ad X d ) ∈ GA1 (R/p),
so a2 , . . . , ad = 0 and a1 6= 0, since R/p is a domain. Since ∩p = {0} we must
have a2 , . . . , ad = 0 and a1 ∈ R∗ , so F = (a0 + a1 X). This shows:
GA1 (R) = Af1(R) for R reduced.
p
If R is not reduced the above shows a2, . . . , ad ∈ Nil (R) = {0} and
a1 ∈ R∗ .
10. ELEMENTARY AND TAME AUTOMORPHISMS 7
Conversely (exercise), if F = (a0 + a1X + · · · + ad X d ) with a2, . . . , ad ∈

Nil (R) and a1 ∈ R∗ , then F ∈ GA1 (R).
10. Elementary and tame automorphisms

• Elementary automorphisms: An automorphism F is called elementary
(or de Jonquière) if it has the form
F = (X1 , . . . , Xi−1 , Xi + f, Xi+1 , . . . , Xn )
where f ∈ R[X1 , . . . , Xi−1 , Xi+1 , . . . , Xn ].
Observe:
(1) F −1 is obtained by replacing “+” by “−”.
(2) Such an F can be viewed as an element of Tr1(R[n−1] ) (by extension
of scalars).
(3) Letting Ti,k be the transposition which switches Xi and Xk , then
Ti,k ◦ F ◦ Ti,k is elementary in the k th position.
Denote by EAn (R) the subgroup generated by all elementary automor-
phisms.
If F (as above) lies in GLn , then we must have
f = a1X1 + · · · + ai−1 Xi−1 + ai+1 Xi+1 + · · · + an Xn
corresponding to the matrix
 
0 0 ··· 0
 .. .. .. .. 
 Ii−1 . . . .
 
 0 0 ··· 0
 
a1 · · · ai−1 1 ai+1 ··· an 
 
0 ··· 0 0 
. . .. .. 
 .. .. . . In−i 
0 ··· 0 0
which factors in GLn as ei1(a1) · · · î · · · ein (an ) ∈ En (R).
Exercise: EAn (R) ∩ GLn (R) = En (R).
Fact from K-theory: [GL∞ , GL∞ ] = E∞ .
( K1 (R) = GL∞ (R)ab = GL∞ (R)/E∞ (R) ).
Question: Is [GA∞ , GA∞ ] = EA∞ ?
Theorem 1.4 (Connell). [GA∞ , GA∞ ] is the normal closure of EA∞ .

This is proved in [17].
• Tame automorphisms: The subgroup of tame automorphisms is defined

as
Tn (R) = h GL n (R) , EAn (R) i
These are all the automorphisms one can “naively” generate, for R a
domain.
Note: T1 (R) = GA1 (R) ⇐⇒ R is reduced.
We will see that not all automorphisms are tame, even for R a domain.
(nontrivial)
In fact we’ll see that, for R a domain,
T2 (R) = GA2(R) ⇐⇒ R is a field.
(⇐ is the Jung - van der Kulk Theorem, which will be proved in §12.)
By observation (3) above, Tn is generated by GLn and all automorphisms
which are elementary in the ith position, for a fixed i.
Exercise: Let Ton = Tn ∩ GAon . Then Ton is generated by GLn and all
origin-preserving elementary automorphisms.
Note: Regarding tameness, the base ring matters when restriction of
scalars is possible. The “Cohn matrix” C is in GL2(R[X, Y ]) hence lies in
T2 (R[X, Y ]). But it is not likely that C ∈ T4 (R). (This is unknown.)
However, C becomes tame in GA5 (R). This is because the matrix
 
1 + XY Y2 0
 −X 2 1 − XY 0
0 0 1
is a product of elementary matrices, i.e., lies in E3 (R[X, Y ]). But an elemen-
tary matrix over R[X, Y ] is an elementary automorphism over R.
So
C [1] ∈ E3(R[X, Y ]) ⊆ T5 (R).
• Triangular automorphisms: An automorphism F = (F1, . . . , Fn ) is
called triangular if (contravariantly) it preserves the containments
R[X1 ] ⊆ R[X1 , X2 ] ⊆ · · · ⊆ R[X1 , . . . , Xn ].
For R reduced this is equivalent to each Fi having the form ui Xi +f (X1 , . . . , Xi−1 ),
for i = 1, . . . , n.
Such automorphisms form a subgroup, which we will denote by BAn (R).
Note that it depends on an ordering of the variables.
Proposition 1.5. For R reduced we have
Tn (R) = h Afn (R), BAn (R) i
11. STABLE TAMENESS 9
The proof is an easy exercise.
11. Stable tameness

Clearly Tn (R) ,→ Tn+1 (R) by the stability homomorphism, so we can
form the direct limit T∞ (R) ⊆ GA∞ (R).
We say F ∈ GAn is stably tame if F ∈ T∞ , equivalently, if F [m] ∈ Tn+m
for some m ≥ 0.
Example of Nagata: Let R be a domain, not a field, and let t be any
non-zero non-unit in R. Let K = frac (R). Let
1
A = (X, Y + X 2 ), B = (X + t2Y, Y )
t
and let
N = A ◦ B ◦ A−1
2 !
1 1 1 1
= X + t2 (Y − X 2 ), Y − X 2 + X + t2 (Y − X 2 ) .
t t t t
We have:
N ∈ GA2 (R) (overring principle),
N ∈ T2 (K),
/ T2 (R) (we’ll see why later).
N∈
Moreover, N is stably tame over R. In fact:
N [1] ∈ T3 (R).
Specifically, letting
A = (X, Y, Z + (tY − X 2 )),
B = (X − tZ, Y, Z)
we have
N [1] = ABA−1B −1 .
However, if R = k[t] a polynomial ring, then
N∈
/ T3 (k)
by restriction of scalars. This will be discussed later.
Question: Is T∞ (R) = GA∞ (R) ?
#$
*'#+)!,
%
!"
&'!$()#"(
Figure 1. The triangle ∆ in the XY -plane
12. The case n = 2, R a field
Theorem 1.6 (Jung, van der Kulk). For k a field (of any characteristic),
T2 (k) = GA2 (k).
The original proofs of this theorem can be found in [43] and [50], the
latter extending the result from characteristic zero to arbitrary characteristic.
We will outline a proof given by Makar-Limanov,
P as modified by Dicks [24].
Given (F, G) ∈ GA2 (k) we write F = αij X i Y j and set m = degX F ,
n = degY F . We view the monomials X i Y j as elements of the free abelian
group hX, Y i which we identify with Z2 via X i Y j ↔ (i, j). We define
supp F = {X i Y j | αij 6= 0}.
and note that supp F lies within the rectangle Γ having vertices 1, X m , X m Y n
and Y n , which contains the triangle ∆ determined by 1, X m , Y n (Figure 1).
12. THE CASE n = 2, R A FIELD 11
If mn ≤ 1 we have F is linear and in this case it is easy to see that (F, G)

is tame. So we assume mn > 1.
Our goal will be to show:
Proposition 1.7. We have
(1) X m , Y n ∈ supp F ,
(2) supp F ⊆ ∆, and
(3) m | n or n | m.
The theorem can be derived from this as follows: Write n = `m (by 3)
and note that n = deg F (by 1 and 2). Set E = (X − α0n α−1 `
m0 Y , Y ) and
−1 `
note that the first coordinate of (F, G) ◦ E is F (X − α0n αm0 Y , Y ) which has
lower Y -degree. (The term α0n Y n gets cancelled and no higher powers of Y
are introduced.) Thus n is lowered and the theorem is proved by induction.
Let
m n
m0 = , n0 =
gcd(m, n) gcd(m, n)
which are coprime; choose integers s, t ≥ 0 such that sm0 − tn0 = 1. Then let
U, V ∈ hX, Y i ⊆ k(X, Y ) be defined by
0
Xm Ys
U = n0 , V = t.
Y X
s n0 t m0
Then X = U V and Y = U V , so hX, Y i = hU, V i and we have the
birational containment k[X, Y ] ⊆ k[U, V ]. Therefore we can write
X
F = νij U i V j .
The change from X, Y to U, V amounts to a change of basis in Z2 , and we
have supp F = {U i V j | νij 6= 0}.
Note that in the U V -plane, Γ is now a parallelogram contained in the
first quadrant. The diagonal of Γ, which is the top side of ∆, is parallel to
the U -axis (Figure 2).
At this point it is expedient to consider the embedding of k[U, V ] in the
ring K[[V −1 ]][V ], where K = k(U ), which is denoted by K((V −1 )). We have
containments:
k[X, Y ] ⊆ k[U, V ] ⊆ k(U, V ) = k(X, Y )
⊆ K((V −1 )).
An element Q of K((V −1 )) can be written uniquely as
X
d
Q= qi (U )V i
i=−∞
& $
#$%& ' (
! "
Figure 2. The triangle ∆ in the U V -plane.
with qi (U ) ∈ K = k(U ) and qd (U ) 6= 0.

We define |Q| to be the leading V -summand qd (U )V d of Q.
Embedding K = k(U ) in k[[U −1 ]][U ] = k((U −1 )), we can write
X
e
qd (U ) = µj U j
j=−∞
with µj ∈ k and µe 6= 0, and we define kQk to be µe U e V d . One easily verifies

that
| | : K((V −1 ))∗ −→ K ∗ × hV i,
k k : K((V −1 ))∗ −→ hU, V i
are both group homomorphisms. Moreover, both can be viewed as retractions.
Both target groups will be viewed as subgroups of k(U, V )∗ .
The proof will now be done in 4 steps. Note that the proposition is con-
tained in the assertions below.
Step 1 There exist A, B ∈ K ∗ × hV i such that |F | = λA` , where λ ∈ k ∗,
and ` ∈ Z+ , and X, Y ∈ k[A±1, B].
Step 2 There exist W, Z ∈ hU, V i such that

• hW i = hkAki or hW i = hkAk, kBki,
• X, Y ∈ moniod [W ±1, Z].
Step 3 We have
• X m , Y n ∈ supp F ,
• supp F ⊆ ∆,
• kF k = X m ,
• hW i = hXi (i.e. W = ±X).
Step 4
• If hW i = hkAki then m | n.
• If hW i = hkAk, kBki then n | m.
Step 1 There exist A, B ∈ K ∗ × hV i such that |F | = λA` , where λ ∈ k ∗,
and ` ∈ Z+ , and X, Y ∈ k[A±1, B].
Proof: In the group (K ∗ × hV i) / k ∗ , which is torsion-free, there is a maxi-
mal cyclic subgroup containing the top summand |F | of F , since degV F > 0.
Let A ∈ K ∗ × hV i be the lift of a generator. Possibly replacing A by A−1, we
have |F | = λA` with ` ∈ Z+ as desired.
We will show that for any F, G ∈ K((V −1 )) with |F| = λA` , ` ∈ Z+ , there
is a B ∈ K ∗ × hV i such that |k[F±1 , G]× | ⊆ k[A±1 , B].
This will prove Step 1, since X, Y ∈ |k[F ±1, G]× |.
The proof is by induction on `, the case ` = 0 being vacuous.
We now define a sequence G1 , G2, . . ., possibly finite, as follows: G1 = G.
If Gi is defined and if |Gi | = λi |F|ni for some λi ∈ k ∗ , ni ∈ Z, then let
Gi+1 = Gi − λi Fni ; otherwise terminate the sequence at Gi .
Since degV G1 > degV G2 > · · · , the sequence G1, G2 , . . . has the limit
G ∗ = G − λ1 F n 1 − λ2 F n 2 − · · ·
in K((V −1 )).
If G∗ = 0 (which is the case if the sequence is infinite), then k[F±1 , G] ⊆
k((F−1 )), so
±1 ×
k[F , G] ⊆ k((F−1 ))∗ ⊆ k[ |F|±1 ] ⊆ k[ |A|±1]
and any B will work.

Thus we may assume G∗ 6= 0, so the sequence is finite and k[F±1 , G] =
±1
k[F , G∗ ].
If |F| and |G∗| are algebraically independent over k, it is easy to see that
|k[F±1 , G∗ ]× | ⊆ k[|F|±1 , |G∗ |], so we can take B = |G∗ |.
This leaves the case where |F| and |G∗ | are algebraically dependent over
the field k.
Letting c = degV F, d = degV G∗ , then |F|d , |G∗|c are algebraically de-

pendent of the same degree. Note that |F|d /|G∗ |c lies in K (since its V -
degree is 0), and is algebraic over k (since |F|d is transcendental over K and
k( |F|d /|G∗ |c , |F|d ) has transcendence degree 1 over k), hence lies in k (since
k is algebraically closed in K = k(U ) ). So |G∗ |c ≡ |F|d ≡ A`d (mod k ∗ ). Since
the image of A in (K × hV i)/k ∗ generates a maximal cyclic subgroup, we
must have c | `d and |G∗| = µAb where b = `d/c, µ ∈ k ∗ . (So the top form
of G∗ is essentially a power of A as well.) But by the definition of G∗ as the
termination of the sequence G1 , G2 , . . . we have ` - b (in particular ` > 1) and
we can write b = `q + r with 0 < r < `.
Let H = G∗ /Fq . Then |H| = νAr , ν ∈ k ∗ , and induction applies to the
pair (H, F), since r < `. So, there exists B ∈ K ∗ × hV i such that
(1) |k[H±1, F]× | ⊆ k[A±1, B].
The induction is now completed by noting that
|k[F±1 , G]× | ⊆ |k[F±1 , H]×| ⊆ |k[F, H]×| h|F|i
⊆ |k[F, H]×| hAi
⊆ k[A±1, B] (by (1)).
This completes Step 1.
Step 2 There exist W, Z ∈ hU, V i such that
• hW i = hkAki or hW i = hkAk, kBki,
• X, Y ∈ moniod [W ±1, Z].
Proof: We have kAk, kBk ∈ hU, V i, and either they are independent, i.e.,
generate a free abelian group of rank two, or they are dependent, i.e., generate
a free cyclic group.
If kAk, kBk are independent, then we must have the containment
kk[A±1, B]×k ⊆ moniod [kAk±1, kBk] .
Since X, Y ∈ k[A±1, B]× , we can take W = kAk, Z = kBk.
If kAk, kBk are dependent, take W to be a generator of hkAk, kBki, so
that kAk = W i , kBk = W j , W = kAkckBkd . Hence
(2) ci + dj = 1.
Then kAkj = kBki = W ij , so there exists a unique number µ ∈ k ∗ such that
kAj − µB i k =
6 W ij . (Note that A, B are algebraically independent (since
X, Y ∈ k[A , B]), so we cannot have Aj − µB i = 0.)
±1
Set Z = kAj − µB i k, and observe that Z and W ij must have the same
V -degree (since A and B are both V -homogeneous). Therefore Z and W ij
are independent, so Z and W are independent.
Let A0 = Ac B d , B 0 = Aj B −i − µ. From (2) above it follows that
k[A±1, B ±1 ] = k[A0±1, (B 0 + µ)±1 ]
(since A = (A0)i (B 0 + µ)d , B = (A0)j (B 0 + µ)−c ). So
kk[A±1,B ±1 ]× k = kk[A0±1, (B 0 + µ)±1 ]× k
⊆ kk[A0, B 0 + µ]× k × hW i (see below)
= kk[A0, B 0]× k × hW i
⊆ moniod [W ±1 , Z].
To justify the second line, note that kA0k = W and kB 0 + µk = 1. Let
H ∈ k[A0±1, (B 0 + µ)±1 ]× and write H = A0M (B 00 + µ)N h(A0 , B 0 + µ) where
h is a polynomial. Then
kHk =kA0kM kB 0 + µkN kh(A0, B 0 + µ)k
= W M kh(A0, B 0 + µ)k.
Thus we have X, Y ∈ moniod [W ±1, Z], and hW i = hkAk, kBki, complet-
ing Step 2.
Step 3 We have
• X m , Y n ∈ supp F ,
• supp F ⊆ ∆,
• kF k = X m ,
• hW i = hXi (i.e. W = ±X).
Proof: By Step 2, kAk = W i , and we can replace W by W −1 to assume
i ≥ 0. By Step 1, |F | = λA` and this implies kF k = kAk` = W i` . Now
kF k ∈ moniod [X, Y ] (being in the parallelogram Γ), and this forces W to
be in Γ as well; in particular, W ∈ moniod [X, Y ].
Now note that since X, Y ∈ moniod [W ±1 , Z], we must have X and Y
lying in the same half plane determined by W , hence W lies on one of the
lower two sides of Γ, which says W , and hence kF k, is a power of X or a power
of Y . But supp F meets the X and Y axes only within ∆, so kF k = X m or
Y n . This implies supp F ⊆ ∆, since any monomial M in Γ − ∆ would have
kMk > kF k.
Also, it is not possible that kF k = Y n as this would say there are no other
points in supp F along the line segment joining Y n and X m , contradicting
the fact that m = degX F . Therefore kF k = X m .
We must have Y n ∈ supp F , as it is the only point in ∆ having Y -degree

n.
Finally, the facts that X is a positive power of W , and that X ∈
moniod [W ±1, Z], with Z and W being independent (by virtue of the fact
that X, Y ∈ moniod [W ±1 , Z]), force W = X. This completes Step 3.
Step 4
• If hW i = hkAki then m | n.
• If hW i = hkAk, kBki then n | m.
Proof: In the first case we have hkAki = hW i = hXi so kAk = X. This
says kAk = ±X, but the facts that kF k = kAk` (from Step 1) and kF k = X m
show kAk = X and ` = m. Hence |F | = λAm , from Step 1. The fact that
A ∈ k(U )[V ] ⊆ k(X, Y ) and that |F | = λAm yield A ∈ k[X, Y ]. We conclude
n = degY F = m degY A.
In the second case we have hkAk, kBki = hW i = hXi then, since X =
0
U s V n , we have n0 Z = degV (hXi) = degV (hkAk, kBki) = degV (hA, Bi),
0
because A and B are V -homogeneous. Since Y ∈ k[A±1, B] and Y = U t V m ,
we have m0 = degV Y ∈ n0 Z, so n0 | m0, which implies n | m.
13. Structure of GA2 (k), k a field

Theorem 1.8 (van der Kulk). For k a field, GA2(k) has the amalgamated
free product decomposition
GA2 (k) = Af2(k) ∗B2 (k) BA2(k)
Before giving the proof we will review some generalities about amalga-
mated free products.
Given groups A, B, C with injective homomorphisms ι1 : B → A, ι2 :
B → C (so we view B as a subgroup of both A and C) we can form the
amalgamated free product H = A ∗C B, which can be realized as the free
product of A and C modulo the relations {ι1 (b) = ι2 (b) | b ∈ B}.
Normal form. Given sets of nontrivial left coset representatives A of A
mod B and C of C mod B, respectively, then an element h ∈ H has a unique
factorization (normal form) h = ba1c1 a2c2 · · · ar (cr ) or h = bc1a1 c2a2 · · · cr (ar )
with r ≥ 0, ai ∈ A, cj ∈ C, b ∈ B.
Now suppose G is a group with subgroups A and C, with B = A ∩ C.
There is an induced map ϕ : A ∗B C → G. In this situation,
(1) ϕ is surjective ⇐⇒ A ∪ C generates G.
(2) ϕ is injective ⇐⇒ whenever
ϕ(ba1c1a2 c2 · · · ar (cr )) = 1
13. STRUCTURE OF GA2 (k), k A FIELD 17
or
ϕ(bc1a1c2 a2 · · · cr (ar )) = 1
(with ai , cj , b as before) then r = 0 and b = 1.
Now let R be a domain, and set G = T2 (R), A = Af2 (R), C = BA2 (R).
Then B = A ∩ C is the group Bf2 (R) consisting of automorphisms of the
form
(uX + a, vY + bX + c)
with u, v ∈ R∗ , a, b, c ∈ R. As before, let H = A ∗B C and consider the map
ϕ : A ∗B C → T2 (R).
This is surjective, by (1). Injectivity follows from (2) and the following
lemma:
Lemma 1.9. Let R be a domain. Given
A1 , . . . , Ar ∈ Af2 (R) − Bf2(R),
C1, . . . , Cr ∈ BA2(R) − Bf2 (R),
write
Ci = (ui X + ei , vi Y + fi (X))
with di = deg fi (= deg Ci ). (Note that di ≥ 2 since Ci ∈
/ Bf2 (R).) Then:
(1) Writing
A1 ◦ C2 ◦ A2 ◦ · · · ◦ Cr (◦Ar ) = (P, Q)
we have d2 · · · dr = deg P ≥ deg Q.
(2) Writing
C1 ◦ A1 ◦ · · · ◦ Cr (◦Ar ) = (S, T )
we have d1 · · · dr = deg T > deg S.
Proof. When r = 1 both statements are obvious. Assume r ≥ 2 and
that the lemma holds for r − 1. Let C2 ◦ A2 ◦ · · · ◦ Cr (◦Ar ) = (S 0, T 0), A1 =
aX + bY + s, cX + dY + t). Note that b 6= 0, since A1 ∈ / Bf2 (R). Inductively
we may assume d2 · · · dr = deg T 0 > deg S 0 , by (2). Then the P and Q of (1)
are determined by (P, Q) = A1 ◦ (S 0, T 0), hence
P = aS 0 + bT 0 + s
Q = cS 0 + dT 0 + t
from which it follows (since b 6= 0) that d2 · · · dr = deg P ≥ deg Q, which
proves (1).
For (2), note that (S, T ) = C1 ◦ (P, Q) hence

S = u1 P + e1,
T = v1 Q + f1 (P ),
from which it follows that deg T = d1 deg P = d1 · · · dr and deg S = deg P =
d2 · · · dr . Therefore deg T > deg S since d1 ≥ 2.
14. Structure of T2 (R), R a domain

What we actually have proved is:
Theorem 1.10. For R a domain, T2 (R) has the amalgamated free prod-
uct decomposition
T2 (R) = Af2 (R) ∗B2 (R) BA2(R)
Moreover a close examination of the proof of Lemma 1.9 yields the fol-
lowing properties of tame automorphisms in dimension 2, which provide an
algorithm for determining if a pair F = (P, Q) determines a tame automor-
phism, and if so, for writing F as the composition of linear and elementary
automorphisms:
Proposition 1.11. Let F = (P, Q) ∈ T2 (R), R a domain, and assume
deg P ≤ deg Q. Then
• If deg P = deg Q ≥ 2 then there exists A ∈ GL2 (R) such that
A ◦ F = (S, T ) with deg S < deg T .
More strongly, the top forms of P and Q have the form c(eL)` ,
d(eL)` , respectively, with c, d, e ∈ R, e 6= 0, and (c, d) unimodular
(which means the row (c, d) can be completed to an invertible matrix
over R, or, equivalently, cR+ dR = R), and L = aX + bY with (a, b)
unimodular in R.
• If deg P < deg Q then there exist s ∈ R and ` ≥ 2 such that deg(Q −
sP ` ) < deg Q. Hence letting E = (X, Y −sX ` ) we have deg(E ◦F ) <
deg F .
More strongly, the top forms of P and Q have the form eLm ,
` `m
se L , respectively, where L is as above.
Corollary 1.12. For R a domain we have
T2 (R) = GA2 (R) ⇐⇒ R is a field.
Proof. The implication ⇐ is the Jung-van der Kulk Theorem. The con-
verse follows from the fact that the Nagata automorphism N (see §11) is not
15. THE CASE n = 3 19
in T(R), which results from the proposition upon noting that the top forms
of N are tX 2 and tX 4 .
Question: For R a domain (even R a UFD, or R = k [1], or R a DVR), is
GA2 (R) ⊆ T∞ (R)? (In other words, are elements of GA2(R) stably tame?)
15. The case n = 3
In 2002 Shestakov and Umirbaev proved this landmark theorem [69]:
Theorem 1.13. For k a field of characteristic zero, T3 (k) is properly

contained in GA3(k).
More strongly, they proved
Theorem 1.14. Considering the containment of GA2 (k [1] ) ⊆ GA3(k) by

restriction of scalars, we have T3 (k) ∩ GA2 (k [1]) = T2 (k [1] ). In particular, the
Nagata automorphism N is not in T3(k).
The proof is difficult and not widely understood.
CHAPTER 2
Perhaps the Jacobian Conjecture is simple
Arno van den Essen
Introduction
In his Nobel lecture on December 11, 1965, the physicist Richard Feyn-
man, after having given various different descriptions of electrodynamics,
came to the following conclusion:
“Perhaps a thing is simple if you can describe it fully in sev-
eral different ways, without immediately knowing that you are
describing the same thing.”
It is the aim of this paper to show that the Jacobian Conjecture un-
doubtedly satisfies the hypothesis of Feynman’s statement, by giving five
completely different, fully equivalent desciptions of it. Only the future will
tell if indeed the Jacobian Conjecture is simple!
1. The classical formulation

Let F := (F1 , . . . , Fn ) : Cn → Cn be a polynomial map, i.e. each Fi
belongs to the polynomial ring in n variables over C, which we denote by
C[n] or C[x1, . . . , xn ] or simply by C[x]. Such a map F is called invertible
if there exists a polynomial map G := (G1 , . . . , Gn ) : Cn → Cn such that
G ◦ F = IdCn , i.e. Gi (F1, . . . , Fn ) = xi for all i.
It can be shown that in this case also Fi (G1 , . . . , Gn ) = xi for all i. Fur-
thermore, invertible polynomial maps correspond one-to-one with C-automor-
phisms of the ring C[x]. This correspondence is given by the following formula
F 7→ (F ∗ : g 7→ g(F )).
So describing invertible polynomial maps from Cn to Cn is the same as de-
scribing C-automorphisms of C[x]. Sometimes we use F to denote the n-
tuple (F1, . . . , Fn ) and sometimes we also use it to denote the corresponding
C-automorphism F ∗.
21
22 2. PERHAPS THE JACOBIAN CONJECTURE IS SIMPLE
Using the chain rule one deduces that, if F is invertible, then
(J G)(F (x))J F = In ,

where JF denotes the Jacobian matrix ∂F i
∂xj
of F . Taking determi-
1≤i,j≤n
nants gives
det((J G)(F (x))detJF = 1,
which implies that detJF ∈ C∗ .
Now one can wonder if the converse is true. This problem became famous
under the name Jacobian Conjecture:
Jacobian Conjecture. For each n ≥ 1 the following statement, denoted by

JC(n, C), holds: if F : Cn → Cn is a polynomial map with detJ F ∈ C∗ , then
F is invertible.
The conjecture was first formulated by O.Keller, in [48] 1939, for the
special case that the polynomials Fi have integer coefficients and became
famous as “Keller’s Problem”. However it can be shown that Keller’s problem
is equivalent to the Jacobian Conjecture ([27], Chapter 1).
Apart from the trivial case n = 1, the conjecture remains open for all
n ≥ 2, inspite of the efforts of many mathematicians during six decades.
More recently the conjecture appeared as “problem 16” on a list of 18 famous
open problems in the paper Mathematical problems for the next century by
the fields medalist Steve Smale [70]. For an extensive history of the conjecture
and its relations with many other open problems, we refer the reader to [9]
and [27].
2. The Rolle formulation

One of the first theorems one meets in a calculus course is Rolle’s theorem,
which asserts that if a differentiable function F on R takes on the same value
at real numbers a and b with a < b, then its derivative is zero at some at
some c with a < c < b.
The main result of this section asserts that if one replaces “a differentiable
function F on R” by “a polynomial map F : Cn → Cn ” and “the derivative
of F at c ∈ R” by “F 0(c), for some c ∈ Cn ”, where F 0(c) := detJ F (c), then
one gets the first equivalent description of the Jacobian Conjecture. More
precisely
2. THE ROLLE FORMULATION 23
Theorem 2.1. The n-dimensional Jacobian Conjecture is equivalent to

the following statement:
(R) If F : Cn → Cn is a polynomial map with F (a) = F (b) for some a 6= b
in Cn , then F 0(c) = 0 for some c ∈ Cn .
Using the fact that a non-constant polynomial has a zero in Cn , one
readily verifies that statement (R) is equivalent to: if detJ F ∈ C∗, then F :
Cn → Cn is injective. So Theorem 2.1 follows from the following remarkable
result.
Theorem 2.2. An injective polynomial map F : Cn → Cn is invertible.
In fact, Theorem 2.2 is a special case of the following more general result
of Cynk and Rusek (see [20] or [27]): if V is an algebraic variety over an
algebraically closed field of characteristic zero, then any injective endomor-
phism of V is an automorphism, in other words any injective endomorphism
of V has an inverse which is also a morphism of algebraic varieties.
On the other hand, for non-algebraically closed fields a similar result
as Theorem 2.2 does not hold. Namely the map F : R → R defined by
F (x) = x3 is even bijective, however its inverse is not a polynomial map. Also
one cannot extend Theorem 2.2 to holomorphic mappings, since there exist
examples, due to Fatou and Bieberbach, of injective holomorphic mappings
F : C2 → C2, even with detJ F = 1, such that C2 minus F (C2) contains a
non-empty open set!
Before we give a sketch of a proof of Theorem 2.2, we will derive a nice
consequence of it.
Proposition 2.3. (Wang) Let F : Cn → Cn be a polynomial map with
detJF ∈ C∗ . If degFi ≤ 2 for all i, then F is invertible.
Proof. By Theorem 2.2 it suffices to prove that F is injective. So assume
F (a) = F (b), for some a and b in Cn which are not equal. Define
G(x) = F (x + a) − F (a).
Then degGi ≤ 2 for all i. Furthermore G(0) = 0 and, putting c = b − a
we have G(c) = 0 and c is non-zero. Observe that JG(x) = J F (x + a), so
detJG ∈ C∗ . Now write G = G(1) + G(2), its decomposition in homogeneous
components, and consider G(tc) = tG(1)(c) + t2G(2) (c). Differentiating gives
d
G(1) (c) + 2tG(2) (c) = G(tc) = (J G)(tc).c 6= 0 for all t ∈ C
dt
since detJG ∈ C∗ and c is non-zero. Finally, substituting t = 1/2 gives
G(c) 6= 0, which contradicts G(c) = 0. So F is injective.
Remark 2.4. Although Proposition 2.3 looks rather far away from the
general case, it was shown in [9] and [78] that it suffices to investigate the
Jacobian Conjecture “only” for all polynomial maps F with degFi ≤ 3 for all
i, in each dimension. So one fails to settle the conjecture by just one degree!
Now let’s turn to the proof of Theorem 2.2. The proof we will give below
consists of the following two parts.
1) First we show that an injective polynomial map F : Cn → Cn is automat-
ically surjective.
2) Then we deduce that the inverse, which by 1) exists as a map of sets, is
in fact a polynomial map.
So let’s look at the first statement. The proof we will present now uses
a very useful technique, namely solving a problem over C, by reducing it to
a problem over a finite field. To be more precise, we will show that every
injective polynomial map F : Cn → Cn is surjective, by reducing it to the
obvious statement that every injective (polynomial) map F : k n → k n is
surjective, in case k is a finite field!
The proof is based on the following general fact (see [27], 4.1.5)
Proposition 2.5. Let A be a finitely generated Z-algebra which is a
subalgebra of some Q-algebra. Then for almost all prime numbers p there
exists a maximal ideal m of A, containing p, such that A/m is a finite field.
Lemma 2.6. Every injective polynomial map F : Cn → Cn is surjective.
Proof. Let F : Cn → Cn be a polynomial map.
i) Then, F is injective if and only if for all (a, b) ∈ Cn × Cn the equations
F1 (a) − F1(b) = 0, . . . , Fn(a) − Fn (b) = 0
imply that
a1 − b1 = 0, . . . , an − bn = 0.
In other words, F is injective if and only if every zero (a, b) ∈ C2n of the ideal
I := (F1(x) − F1(y), . . . , Fn (x) − Fn (y))
in C[x, y] is a zero of the ideal J := (x1 − y1 , . . . , xn − yn ), or equivalently
(by the Nullstellensatz), there exist m ∈ N and aij (x, y) in C[x, y] such that
for all i
X
(1) (xi − yi )m = aij (x, y)(Fj (x) − Fj (y)).
j
2. THE ROLLE FORMULATION 25
ii) Suppose that F is not surjective. Then there exists a point (c1 , . . . , cn ) ∈
Cn which does not belong to F (Cn ), in other words such that the system
F1(x) − c1 = 0, . . . , Fn(x) − cn = 0
has no solution in Cn . By the Nullstellensatz this is equivalent to: there exist
a1(x), . . . , an (x) in C[x] such that
X
(2) 1= aj (x)(Fj (x) − cj ).
j
iii) Now let S be the subring of C generated over Z by all coefficients ap-
pearing in the equations (1) and (2). So S is a finitely generated Z-algebra.
By Proposition 2.5 it then follows that there exists a maximal ideal m of S
such that k := S/m is a finite field. Now passing from S to k we get k n and
an induced map F0 : k n → k n . From (1) it follows (by reduction modulo m)
that F0 is injective and by (2) we get that c does not belong to F0(k n ). So
F0 : k n → k n is not surjective. However an injective map from a finite set to
itself (in our case F0 : k n → k n is such a map) is automatically surjective. So
F0 is surjective, and we have a contradiction!
Proof. (of Theorem 2.2) By Lemma 2.5 F is surjective, so by Theorem
2.7 below, using that F is injective, so #F −1(u) = 1 for almost all u in
Cn , we get that C(x1, . . . , xn ) = C(F1 , . . . , Fn ). So there exist polynomials
gi , hi ∈ C[y] such that
xi = gi (F1, . . . , Fn )/hi (F1, . . . , Fn ) and g.c.d.(gi , hi ) = 1 for all i.
Now let 1 ≤ i ≤ n be fixed and suppose that hi 6∈ C∗ . Then there exists b

in Cn such that hi (b) = 0. Since F is surjective we have that b = F (a) for
some a in Cn . From hi (F (x))xi = gi (F (x)) we deduce that gi (b) = 0. So each
zero of hi is a zero of gi . In other words the zero-set of hi is contained in
the zero-set of gi , whence it follows from the Nullstellensatz that for some
positive integer m gim belongs to the ideal (hi ). Since hi is not a constant, it
contains some irreducible factor and hence it follows that g.c.d (hi , gi ) 6= 1,
a contradiction. So hi ∈ C∗ . Consequently xi ∈ C[F1, . . . , Fn ]. This holds for
all i, whence C[x] = C[F ]. This implies that F is invertible.
To conclude this section we give the result we used in the proof above
Theorem 2.7. Let F : Cn → Cn be a surjective polynomial map, then
there exists a non-zero polynomial g in C[x], such that for every point u
in U , the complement of the zero-set of g, the following holds: #F −1 (u) =
the degree of the field extension of C(x1, . . . , xn ) over C(F1, . . . , Fn ).
3. The degree of the inverse

To give the next equivalent description of the Jacobian Conjecture we
need to go back to 1980. Then Stuart Wang showed (in [77]) that the Ja-
cobian Conjecture is true in case F : k n → k n is a polynomial map with
degF := max degFi ≤ 2, where k is a field of characteristic not equal to 2.
In fact, he even showed this result in case k is a UFD with 2 6= 0. At the end
of his paper he makes the following Degree Conjecture: if k is a UFD with
2 6= 0 and F a polynomial automorphism of k[x] over k with degF ≤ 2, then
degF −1 ≤ 2n−1 .
This conjecture was remarkable at that time, since it contrasted with an
earlier conjecture of Sathaye, who, as Wang writes, states that the degree of
the inverse of a polynomial automorphism “in general” is not bounded.
Wang’s conjecture was proved in the field case around 1980 by Rusek
and Winiarski (and published some time later in [60]) and simultaneously
by Gabber (see [9]). In fact they proved the following result.
Proposition 2.8. (Rusek, Winiarski, Gabber) Let k be any field and F

a k-automorphism of k[x]. Then
degF −1 ≤ (degF )n−1 .
Using standard arguments from commutative algebra, this degree esti-

mate can be easily extended to the case that k is a reduced ring, i.e. a ring
without nilpotent elements (see [27], 2.3.4]). So the next question is: what
happens if k has nilpotent elements?
It turns out that in this case we have a completely different situation,
as can be seen from the following example, which shows in particular that
Sathaye wasn’t wrong after all!
Example 2.1. Let n = 1 and m ≥ 2. Put Cm = C[T ]/(T m) and ε = T .

Then εm = 0 and εm−1 6= 0. Define F (x) = x + εx2. Then degF = 2, however
F is a Cm -automorphism of Cm [x] with degF −1 = m. In other words the
degree of the inverse of F can be arbitrarily large, although the degree of F
is 2.
To see this we compute the inverse of F . Therefore we just need to solve

x from the quadratic equation F (x) = y, in other words we have to solve x
from the equation
εx2 + x = y.
3. THE DEGREE OF THE INVERSE 27
One easily finds that

(−1 + (1 + 4εy) 2 ) X 1/2
1 m
x= = 2 (4ε)i−1 y i .
2ε i=1
i
So indeed F is a Cm -automorphism of Cm [x] and the degree of the inverse of
F is equal to m.
At this moment the reader may be wondering why on earth are we con-
sidering the degree of the inverse of a polynomial automorphism, since for
studying the Jacobian Conjecture one has to study polynomial maps whose
Jacobian determinant is equal to a non-zero constant and for which we have
to prove that F is an automorphism. So why should we study the assumption
that F is an automorphism?
The answer is given by the following results. The first one can be found
in [9] and [29].
Theorem 2.9. Let n ≥ 1. If the Jacobian Conjecture is true, then the
following Uniform Boundness statement is true as well:
UB(n). For every d ≥ 1 there exists a positive integer C(n, d) with the follow-
ing property: for every Q-algebra R and every R-automorphism F of R[x],
having detJF =1 and degF ≤ d, the degree of its inverse is bounded by C(n, d)
The point in the Uniform Boundness statement is that one only considers
R-automorphisms F of R[x] having detJF = 1. This restriction rules out our
“bad” example above, whose degree can be arbitrarily large, depending on
the ring R. Namely if F = x+εx2, then detJ F = 1+2εx 6= 1. Apparently, by
Theorem 2.9, the existence of a uniform bound C(n, d) for the degree of the
inverse of an R-automorphism with detJ F = 1, independent of the Q-algebra
R, is a necessary condition for the Jacobian Conjecture to be true.
Now the remarkable point, observed by Bass around 1983 (in [8]), is
that the uniform boundness statement is also sufficient for the Jacobian
Conjecture to be true. In fact a stronger converse, with a very elegant proof,
was obtained by Harm Derksen in 1994 (in [21]):
Theorem 2.10. (Derksen) Let n ≥ 1. Then U B(n) implies JC(n, C),
where U B(n) is the following (weaker) version of the uniform boundness
statement:
U B(n). For every d ≥ 1 there exists a positive integer C(n, d) with the fol-
lowing property: for every m ≥ 1 and every Cm -automorphism F of Cm [x]
with detJF = 1 and degF ≤ d, the degree of its inverse is bounded by C(n, d).
In other words the degree of the inverse of a Cm -automorphism F of Cm [x]

with detJF = 1 is bounded by a constant which does not depend on m.
Proof. Let d ≥ 1 and F be a polynomial automorphism of C[x] with
detJF ∈ C∗ and deg F = d. We may assume that F (0) = 0 and that
JF (0) = In . So detJF = 1. Let G ∈ C[[x]]n be its (unique) formal inverse.
Denote by G(i) the homogeneous component of G of degree i. We will show
that G(l) = 0 for all l > C(n, d).
Therefore we introduce a new variable T and define
F T = T −1F (T x) = x + T F(2) + T 2F(3) + . . . + T d−1F(d)

and
GT = T −1 G(T x) ∈ C[T ][[x]]n.

One easily verifies that detJx F T = detJF (T x) = 1. Furthermore
F T ◦ GT = x = GT ◦ F T
(compositition of formal power series in x). Now let l > C(n, d). Reducing
the previous equality mod T l we obtain that F T is a Cl -automorphism of
Cl [x], whose inverse is equal to
l−1
GT = x + G(2) T + . . . + G(l) T .
Also detJx F T = 1 and degF T ≤ d. So by U B(n) we get
(*) deg GT ≤ C(n, d).
l−1
However, GT = x + G(2) T + . . . + G(l) T . So if G(l) 6= 0 then, using that
l−1
T 6= 0, we get that degGT = l > C(n, d), contradicting (*). So G(l) = 0
for all l > C(n, d). In other words G is a polynomial map, as desired.
So Derksen’s theorem asserts that one has to consider as coefficient rings only
the special Q-algebras of the form Cm , with m ≥ 2 and all Cm -automorphisms
with detJF = 1.
Now one can push this result one step further: namely in [28] and [29]
the author obtained the following reformulation of the Jacobian Conjecture,
which asserts that one “only” needs to consider a very special class of Cm -
automorphisms F of Cm [x] with det J F = 1, namely so-called exponential
automorphisms coming from nilpotent derivations with divergence zero.
Before we can state this result we need to introduce some useful notions.
Let A and R be commutative rings.
3. THE DEGREE OF THE INVERSE 29
Definition 2.11. i) A derivation D on A is an additive map D : A → A

satisfying Leibniz’ rule, i.e. D(a + b) = D(a) + D(b) and D(ab) = aD(b) +
D(a)b for all a and b in A.
ii) Such a derivation is called locally nilpotent if for each a ∈ A there exists
a positive integer m such that Dm (a) = 0.
Example 2.2. Let A = R[x] = R[x1, . . . , xn ]. Then the “usual” partial
∂
derivative ∂x i
is a derivation on A. Furthermore each R-derivation on A, i.e.
a derivation D such that D(r) = 0 for all r in R, is of the form
∂ ∂
D = D(x1 ) + . . . + D(xn ) .
∂x1 ∂xn
∂
The derivation ∂xi
is a locally nilpotent derivation on A.
Why are locally nilpotent derivations useful in the study of polynomial
automorphisms? To answer this question we introduce the exponential map
of a locally nilpotent derivation D by the following formula
X∞
1 i
expD(a) = D (a).
i=0
i!
Observe that this sum is finite, since D is locally nilpotent.
The importance of the exponential map comes from the following result,
whose proof can be found in [27].
Proposition 2.12. Let D be a locally nilpotent derivation on A. Then
expD is a ring automorphism of A. Its inverse is the map exp(−D).
Let D be a locally nilpotent R-derivation on R[x]. Then by Proposition
2.12 F = expD is an R-automorphism of R[x]. Furthermore, it can be shown
that if divD = 0, where
∂ ∂
divD = (D(x1 )) + . . . + (D(xn )),
∂x1 ∂xn
then detJF = 1.
Now we are ready to formulate the improvement of Theorem 2.10 an-
nounced above. It was obtained by the author in [28] (see also [29]), and
asserts that in the formulation of Theorem 2.10 it suffices to study those
Cm -automorphisms of Cm [x] which are of the form expD, where D is a (lo-
cally) nilpotent Cm -derivation of Cm [x] with the property that divD = 0 and
D = 0, where D is obtained from D by reducing its coefficients mod T . In
other words we get
Theorem 2.13. JC(n, C) is equivalent to the following statement: for

every d ≥ 1 there exists a positive integer C∗ (n, d) such that for every m ≥ 2
and every Cm -derivation D of Cm [x] with divD = 0 and D = 0 we have:
if deg expD ≤ d, then deg exp(−D) ≤ C∗(n, d).

4. Zhao’s vanishing conjecture
Recently yet another completely unrecognizable reformulation of the Ja-
cobian Conjecture was obtained by Wenhua Zhao in his paper [80].
Although it is easy to describe his conjecture, we will give it at the end
of this section, since we prefer first to give some indication how it was found.
Starting point is the following classical result, due to Bass, Connell,
Wright and Yagzhev, around 1980 (see [9], [78] and [29]).
Theorem 2.14. (Bass, Connell, Wright/Yagzhev) To study the Jacobian
Conjecture it suffices to study polynomial maps of the form
F = x − H = (x1 − H1 , . . . , xn − Hn )
in all dimensions n, where each Hi is either zero or a homogeneous polyno-
mial of degree 3.
Furthermore, for such maps F the condition detJ F ∈ C∗ is equivalent to J H
being nilpotent.
Then in 2003 Michiel de Bondt and the author (in [13]) improved this
result, by showing that in Theorem 2.14 one may additionally assume that
JH is a symmetric matrix, i.e. J H T = JH. In other words we have
Theorem 2.15. (de Bondt, van den Essen) To study the Jacobian Con-
jecture it suffices to study polynomial maps of the form F = x − H, where
each Hi is either zero or homogeneous of degree 3 and J H is nilpotent. Fur-
thermore one may additionally assume that JH is symmetric.
The condition that JH is symmetric is very restrictive. It implies that
∂i Hj = ∂j Hi for all i, j. By the classical Poincaré lemma it follows that there
∂f
exists a polynomial f ∈ C[x] such that Hi = ∂x i
for all i. So
∂f ∂f
H=( ,..., )
∂x1 ∂xn
the gradient of f , which we denote by ∇f . If furthermore each Hi is either
zero or homogeneous of degree 3, it follows that we can choose f to be
homogeneous of degree 4. So H = ∇f with f homogeneous of degree 4.
Hence
4. ZHAO’S VANISHING CONJECTURE 31
∂ 2f
JH = J (∇f ) = ( )1≤i,j≤n
∂xi∂xj
the so-called Hessian of f , which we denote by Hes(f ). So by Theorem 2.15
the Jacobian Conjecture can be reformulated as follows
Proposition 2.16. The Jacobian Conjecture is equivalent to the follow-
ing statement. For every n ≥ 1 and every homogeneous polynomial f of
degree 4 in C[x] we have: if Hes(f )(= J (∇f )) is nilpotent, then x − ∇f is
invertible.
In his paper [80] Zhao was able to reformulate both the hypothesis
“J (∇f ) is nilpotent” and the statement “x − ∇f is invertible”, of Propo-
sition 2.16 , in terms of the Laplace operator ∆ = ∂12 + . . . + ∂n2 .
First let’s take a look at the nilpotency of the matrix J (∇f ). In general,
if A is an n × n matrix with entries in a field, then A is nilpotent if and
only if T rAm = 0 for all 1 ≤ m ≤ n. Applying this result to the matrix
J (∇f ) = Hes(f ), whose entries belong to the field of rational functions, we
get
Hes(f ) is nilpotent if and only if T rHes(f )m = 0 for all 1 ≤ m ≤ n.
One easily verifies that T rHes(f ) = ∂12(f ) + . . . + ∂n2 (f ) = ∆f . On the
other hand studying the conditions T rHes(f )2 = 0, T rHes(f )3 = 0, etc. is
much more involved. Nevertheless, in [80], Zhao succeeds in expressing these
conditions in terms of the Laplace operator:
Theorem 2.17. (W. Zhao) Let f ∈ C[x]. Then the following statements
are equivalent.
i) Hes(f ) is nilpotent.
ii) ∆m (f m ) = 0 for all 1 ≤ m ≤ n.
iii) ∆m (f m ) = 0 for all m ≥ 1.
So the condition “J (∇f )(= Hes(f )) is nilpotent” is completely under-
stood in terms of the Laplace operator.
Then Zhao continues to study the condition “F = x − ∇f is invertible”.
First he observes that since degf = 4 ≥ 2 it follows that F has a unique
formal inverse G (using the well-known Formal Inverse Function Theorem).
Furthermore JG is symmetric: namely J (F ◦ G) = I, so J F (G)J G = I,
which implies that JG = (JF (G))−1 . Since the inverse matrix of a symmetric
matrix is also symmetric, it follows from the symmetry of J F that also JG
is symmetric.
Using the Poincaré lemma again we obtain that G = x + ∇g for some

g ∈ C[x]. Then Zhao proves the following beautiful theorem.
Theorem 2.18. (W. Zhao) Let f be homogeneous of degree ≥ 3. Then
the formal inverse of x − ∇f is of the form x + ∇g, where g is given by the
formula
X∞
1
g= m
∆m (f m+1 ).
m=0
2 m!(m + 1)!
Now let’s turn to the statement “x − ∇f is invertible”. Since the formal

inverse of x − ∇f is of the form x + ∇g it follows that x − ∇f is invertible
if and only if x + ∇g is a polynomial map.
Now observe that since f is homogeneous of degree 4, either
∆m (f m+1 ) = 0
or it is a homogeneous polynomial of degre 4(m + 1) − 2m = 2m + 4. So each
component of
X∞
g= g2m+4
m=0
1
where g2m+4 = 2m m!(m+1)!
∆m (f m+1 ) is either zero or a homogeneous polyno-
mial of degree 2m + 4. Finally
X
∞
∇g = ∇g2m+4
m=0
and for each i ≥ 1 one has that ∇gi = 0 if and only if gi = 0. So we get
that x + ∇g is a polynomial map if and only if ∇g2m+4 = 0 for all large m,
if and only if g2m+4 = 0 for all large m. So x − ∇f is invertible if and only
if ∆m (f m+1 ) = 0 for all large m. So the condition “x − ∇f is invertible” is
equivalent to the statement
∆m (f m+1 ) = 0 for all large m.
Then, by Proposition 2.16 and Theorem 2.17 the Jacobian Conjecture is
equivalent to
Vanishing Conjecture: (Zhao) For every n ≥ 1 and every homogeneous
polynomial f ∈ C[x] of degree 4, the following statement holds:
if ∆m (f m ) = 0 for all m ≥ 1, then ∆m−1 (f m ) = 0 for all large m.
5. THE DIXMIER CONJECTURE 33
Final remark. The vanishing conjecture has been proved for all n ≤ 5 by
de Bondt and the author in [15], even in case the homogeneous polynomial
f has degree d ≥ 2.
5. The Dixmier Conjecture

To give our last description of the Jacobian Conjecture we consider the
so-called nth Weyl algebra over C. This is the subring of all C-linear operators
on C[x], generated by the multiplication operators xi and the derivations ∂x∂ j .
This ring will be denoted by An (C). So we get
∂ ∂
An (C) = C[x1, . . . , xn , ,..., ]
∂x1 ∂xn
and its generators xi and ∂x∂ j satisfy the following (Heisenberg’s) commutation
relations:
[xi , xj ] = 0, [∂i, ∂j ] = 0, [∂i , xj ] = δij for all i, j.
It is not difficult to prove that An (C) is a simple ring, i.e. that it has no
two-sided ideals, except the zero-ideal and the whole ring (see [12]). Since
the kernel of any C-endomorphism ϕ of An (C) is a two-sided ideal in An (C)
and 1 does not belong to such a kernel, it follows that kerϕ = 0. In other
words any C-endomorphism of An (C) is injective!
Now one may wonder if any C-endomorphism of An (C) is also surjective.
The question was posed for the first time, in case n = 1, by Dixmier in his
paper [23]. This question, generalized to all dimensions, became known as
the Dixmier Conjecture.
Dixmier Conjecture. For any n ≥ 1 the following statement holds:
DC(n, C) Every C-endomorphism of An (C) is surjective (and hence it is a
C-automorphism).
What is the relation between the Dixmier Conjecture and the Jacobian Con-
jecture?
First of all, it was observed by Kac and Vaserstein in the eighties of the
last century that the Dixmier Conjecture implies the Jacobian Conjecture.
More precisely
Proposition 2.19. (Kac, Vaserstein) For each n ≥ 1 DC(n, C) implies
JC(n, C).
Proof. Let F := (F1, . . . , Fn ) : Cn → Cn be a polynomial map with
detJF ∈ C∗. So ((JF )−1)T ∈ Mn (C[x]). Hence, if we define
 ∂
  
∂F1 ∂1
 ..  .
 = ((JF ) )  .. 
−1 T
 .
∂
∂Fn
∂n
∂
then we get derivations ∂F i
on the polynomial ring C[x]. These derivations
have the following properties (see [27], 2.2.3):
∂ ∂ ∂
[ , ] = 0 and [ , Fj ] = δij for all i, j.
∂Fi ∂Fj ∂Fi
So, by defining
∂
ϕ(∂i) = and ϕ(xj ) = Fj
∂Fi
we get a C-endomorphism ϕ of An (C). Since, by assumption DC(n, C) holds,
it follows that ϕ is surjective. In other words
∂ ∂
C[F1, . . . , Fn , ,..., ] = An (C).
∂F1 ∂Fn
P ∂ α
In particular, if g ∈ C[x], then g = aα(F )( ∂F ) , for some aα(F ) ∈ C[F ].
Applying the operator g to the element 1 we get that g = a0 (F ) ∈ C[F ].
So each polynomial of C[x] belongs to C[F ]. In other words C[x] ⊆ C[F ],
which implies that C[x] = C[F ]. So in particular each xi is of the form
Gi (F1, . . . , Fn ), whence F is invertible with inverse G = (G1 , . . . , Gn ).
So the Dixmier Conjecture implies the Jacobian Conjecture. Now the
surprising result, first obtained by Tsuchimoto in [74] and also discussed by
Belov and Kontsevich in [10], is that the converse holds! In other words the
Dixmier Conjecture is just one more, completely unrecognizable, description
of the Jacobian Conjecture! More precisely we have
Theorem 2.20. For each n ≥ 1 we have: JC(2n, C) implies DC(n, C)
and hence the Jacobian Conjecture and the Dixmier Conjecture are equiva-
lent.
The proof which we will give below is due to Adjamagbo and the author.
It was inspired by the paper [10] of Belov and Kontsevich. In our proof we
introduce one more conjecture, the Poisson Conjecture, which also turns out
to be equivalent to the Jacobian Conjecture.
In order to formulate this new conjecture we introduce some notations
and give some definitions. In particular we also define the n−th Weyl algebra
over an arbitrary commutative ring R (in case R = C we recover the definition

given in the beginning of this section).
So let R be a commutative ring with 1 and n a positive integer. The
polynomial ring in x1 , . . . , xn over R is denoted by R[n] or R[x1 , . . . , xn ]. The
n-th Weyl algebra, over R, denoted by An (R), is the associative R-algebra
with 2n generators y1, . . . , y2n and relations
[yi, yi+n ] = 1 for all 1 ≤ i ≤ n and [yi , yj ] = 0
otherwise.
Finally, the n-th canonical Poisson algebra over R is the polynomial ring
[2n]
R endowed with the canonical Poisson bracket {, } defined by
Xn
∂f ∂g ∂f ∂g
{f, g} = ( − ).
i=1
∂x i ∂xi+n ∂x i+n ∂xi
An R-homomorphism ϕ of R[2n] such that {ϕ(f ), ϕ(g)} = {f, g} for all f, g is

called an endomorphism of Pn (R). Now we are able to formulate the Poisson
conjecture.
Poisson Conjecture. For every field K of characteristic zero and every n
the following statement holds:
P C(n, K).Every endomorphism of Pn (K) is an automorphism.
In the last section (in order not to disturb the main line of the proof
of the equivalence of the Dixmier and Jacobian Conjectures) we prove the
following result
Theorem 2.21. If ϕ is an endomorphism of Pn (R) and n! is a unit in
R, then the determinant of the Jacobian matrix of ϕ equals 1.
As an immediate consequence we obtain
Corollary 2.22. JC(2n, C) implies P C(n, C).
Because of this result and Proposition 2.19, it suffices to prove that
P C(n, C) implies DC(n, C) (in order to establish the equivalence of the three
conjectures). Before we start discussing this implication, we need a few more
preparations.
Let ϕ be a C-endomorphism of An (C). Then ϕ is completely determined
by the images ϕ(yi ). The degree of ϕ is by definition the maximum of the
degrees of the ϕ(yi ) (the degree of a non-zero element a of An (C) is the
maximum of the degrees of all monomials cy α, c ∈ C∗ appearing in a).
Then ϕ is surjective if and only if there exist ψ1, . . . , ψ2n in An (C) such
that ϕ(ψi) = yi for all i. If ψi has degree ≤ N , for some N , we say that ϕ is
N-surjective.
Now let’s explain how we are going to prove that P (n, C) implies D(n, C).
The idea is to reduce the proof of this implication to a similar statement
where C is replaced by a ring of the form R/pR, using a technique similar to
the one explained in section two. In order to perform this reduction, we will
rewrite both the hypothesis and the claim in terms of polynomial equations.
Let’s first consider the statement “D(n, C)”, in other words the statement
that every C-endomorphism of An (C) is surjective. In fact we will prove the
following more precise statement:
for every d ≥ 1, every C-endomorphism of An (C) of degree ≤ d is D-
surjective, where D = d2n−1 .
To prove this statement assume the contrary. So let ϕ be a C-endomorphism
of An (C), say of degree d, which is not D-surjective. This means that there
do not exist ψ1, . . . , ψ2n of degree ≤ D in An (C) such that
ϕ(ψ1) − y1 = 0, . . . , ϕ(ψ2n ) − y2n = 0.

So if we consider the universal Weyl algebra elements of degree D, i.e. the
P (i) α (i)
expressions ψiU = |α|≤D cα y , where all cα are different variables and
consider the formal expressions
X
ϕ(ψiU ) − yi := cα(i) ϕ(y1)α1 . . . ϕ(y2n )α2n − yi
|α|≤N
then the coefficients P1 (C), . . . , Ps (C) of all monomials y α appearing in these

expressions (each Pi (C) is a polynomial in the polynomial ring C[C], which
(i)
is generated over C by all coefficients cα ), have no common zero in CC . So
by the Nullstellensatz there exist Q1(C), . . . , Qs (C) in C[C] such that
P
(3) 1 = Qj (C)Pj (C).
Corollary 2.23. Let R be a subring of C containing all coefficients of
all monomials y α appearing in the ϕ(yi ) and all coefficients appearing in all
Qj and Pj . If a is a proper ideal in R (i.e. a is not the whole ring) and ϕ̄
denotes the endomorphism of An (R/a), obtained by reducing the coefficients
of ϕ modulo a, then ϕ̄ is not D-surjective.
Proof. Reduce the equations in (3) modulo a.
Now let’s rewrite the hypothesis P (n, C) in terms of polynomial equa-
tions. To make the situation more finite, we consider for each d ≥ 1 the
statement P (n, C, d), which asserts that the Poisson Conjecture holds for all
endomorphisms of Pn (C) of degree ≤ d. This means that if F = (F1, . . . , F2n )
satisfies degFi ≤ d for all i and {Fi, Fj } = {xi , xj }, then F has an inverse (of
degree ≤ d2n−1 , according to Proposition 2.8.
To rewrite this statement in terms of polynomial equations we consider
the universal polynomial map of degree d in 2n variables x = (x1, . . . , x2n),
i.e.
FiU := (F1U , . . . , F2n
U
),
P (i) (i)
where each FiU := Aα xα with |α| ≤ d and all Aα are different variables.
(i)
Let A := (. . . , Aα , . . .) and denote by C[A] the polynomial ring in A over
C. Consider the polynomials Pij := {FiU , FjU } − {xi , xj } for all 1 ≤ i <
j ≤ 2n and let J := (g1 (A), . . . , gr (A)) be the ideal in C[A] generated by
the coefficients of all monomials xα appearing in all Pij . Then by Theorem
2.21 the canonical image of F U in (C[A]/J )[x]2n has Jacobian determinant 1
and hence, by the formal inverse function theorem, it has a formal inverse in
(C[A]/J )[[x]]2n, represented by some G(A) in C[A][[x]]2n. Let I be the ideal
in C[A] generated by the coefficients in G(A) of all xα with |α| > D := d2n−1
and let h1 (A), . . . , ht (A) be a system of generators of I.
(i)
Proposition 2.24. If P (n, C, d) holds, there exist bj in C[A] and a
positive integer ρ such that
P (i)
(4) hi (A)ρ = j bj (A)gj (A) for all 1 ≤ i ≤ t.
Proof. Let a ∈ CA be a zero of J . Then, since P (n, C, d) holds, the
map F U (A = a) is invertible and its inverse is equal to G(A = a). So by
Proposition 2.8 degG(A = a) ≤ d2n−1 = D. Hence a is a zero of I. So every
zero of J is a zero of I. Then (4) follows from the Nullstellensatz.
As an immediate consequence of Proposition 2.24 we get
Corollary 2.25. Let R be a subring of C containing the coefficients of
(i)
the polynomials hi (A), bj (A) and gj (A). If a is a proper ideal of R and f
an endomorphism of Pn (R/a) of degree ≤ d, then f has an inverse of degree
≤ D(= d2n−1 ).
Proof. (of P (n, C, d) → D(n, C, d).) So assume P (n, C, d) and assume,
as above, that ϕ is a C-automorphism of An (C) of degree d which is not
surjective. This last assumption gives rise to the equations (3). Furthermore,
the assumption P (n, C, d) gives rise to the equations (4).
Let R be the Z-subalgebra of C, generated by n!1 , all coefficients of the
monomials y α appearing in the ϕ(yi ), all coefficients appearing in the hi , gj
(i)
and the bj and all coefficients appearing in all Qj and Pj . Then ϕ is an
R-endomorphism of An (R) and R is a finitely generated Z-algebra without
Z-torsion. By Proposition 2.5 we can choose a prime number p which is not
invertible in R. By reduction modulo pR we get an R/pR-endomorphism ϕ̄
of An (R/pR). By Corollary 2.23 ϕ̄ is not D-surjective. On the other hand
we will show now that the hypothesis that P (n, C, d) holds, implies that ϕ̄
is D-surjective.
To obtain this contradiction we will use the following two theorems. The
first is proved in [5] and [4] and a proof of the second will be given at the
end of this section.
Theorem 2.26. The center of An (R/pR), denoted by Z, is a polynomial
ring in y1p, . . . , y2n
p
over R/pR. Furthermore:
i) ϕ̄(Z) ⊆ Z, so ϕ̄ induces a polynomial map, denoted by ϕ̄pol from the
polynomial ring R/pR[y1p , . . . , y2n
p
] to itself.
ii) If ϕ̄pol is surjective, then ϕ̄ is surjective. More precisely, if ϕ̄pol is N-
surjective for some N ≥ 1, then ϕ̄ is N-surjective.
Theorem 2.27. ϕ̄pol is an endomorphism of Pn (R/pR) (of degre ≤ d).
Now we are able to give
Proof of the surjectivity of ϕ̄. By Theorem 2.27 ϕ̄pol is an endo-
morphism of Pn (R/pR) of degree ≤ d. So it follows from Theorem 2.21, the
hypothesis P (n, C, d) and by reducing the equations (4) in Proposition 2.24
modulo pR, that ϕ̄pol has an inverse of degree ≤ D. It then follows from
Theorem 2.26 ii) that ϕ̄ is D-surjective, which completes the proof.
To prove Theorem 2.27 we give another description of the Poisson bracket
on Pn (R/pR). Put W := R[y1p, . . . , y2n p] and xi := yip for each i.
Lemma 2.28. Let A ∈ W and B ∈ An (R). Then [A, B] ∈ pAn (R).
Proof. Since [ , ] is R-bilinear we may assume that A = (y1p)α1 . . . (y2n p )α2n
and B = y1β1 . . . y2n β2n . Using Leibniz’ rule we may even assume that A = yi p
and B = yj , in which case the result is clear.
Proposition 2.29. Let a, b ∈ R/pR[x1 . . . , x2n ] and A, B ∈ W be such
that a = A(mod pAn(R)) and b = B(mod pAn (R)). Then 1p [A, B] is a well-
defined element of An (R) and
1
{a, b} = [A, B](mod pAn (R)).
p
Proof. Since R has no Z-torsion the first statement follows from Lemma
2.28. To prove the formula, observe that both { , } and [ , ] are bilinear,
antisymmetric and satisfy Leibniz’ rule. Therefore it suffices to show that
1
{xi , xj } = [yip, yj p ](mod pAn (R))
p
for all i < j. If j 6= i + n both sides are zero. So assume j = i + n. Then the
result follows from the following formula
p
1 p 1X (p!)2
(∗∗) [yi+n , yip] = y k y k = −1(mod pZ[yi , yi+n ]),
p p (k!)2(p − k)! i i+n
k=0
which can be found in [23].

Proof. (of Theorem 2.27.) i) First we show that ϕ̄pol is an endomorphism
of Pn (R/pR). Therefore we will show that {ϕ̄pol (xi ), ϕ̄pol(xj )} = {xi , xj } for
all i < j. Since yip ∈ Z it follows that ϕ(yip) = A + pA0 with A in W and A0
in An (R). Similarly ϕ(yjp) = B + pB 0 with B in W and B 0 in An (R). Then
by Proposition 2.29
1
{ϕ̄pol (xi), ϕ̄pol (xj )} = [A, B](mod pAn(R)).
p
Since by Lemma 2.28 both [A, B 0] and [A0, B] belong to pAn (R) we get
1 1 1
[ϕ(yip), ϕ(yjp )] = [A, B]+[A, B 0]+[A0, B]+p[A0, B 0] = [A, B](mod pAn (R)).
p p p
So
1
{ϕ̄pol(xi ), ϕ̄pol (xj )} = [ϕ(yip), ϕ(yjp )](mod pAn (R)).
p
If j 6= i + n the right-hand side is equal to zero and hence equals {xi , xj }.
Finally if j = i + n the right-hand side equals ϕ( 1p [yip, yi+n p
])(mod pAn (R)),
which by (**) is equal to 1 and hence equals {xi, xi+n }.
ii) Finally, the remaining statement that ϕ̄pol has degree ≤ d, follows imme-
diately from the following result.
Lemma 2.30. Let τ be an R/pR-endomorphism of An (R/pR) and τpol the
induced R/pR-endomorphism of the polynomial ring Z. Then τ and τpol have
the same degree.
Proof. Put xi := yip. Then deg τ (yip) = pdegτpol (xi ). Also degτ (yip) =
degτ (yi )p = pdegτ (yi ). So degτpol (xi ) =degτ (yi) for each i, which gives the
desired result.
6. Endomorphisms of canonical Poisson algebras over

commutative rings
Throughout this section R denotes a commutative ring. Let A = R[2n] ,
(e) := (e1 , . . . , e2n ) the standard basis of the free A-module E := A2n and
(e∗) := (e∗1 , . . . , e∗2n ) the dual basis of (e). The canonical symplectic form on
E is the bilinear form given by
X
n
ω := e∗i ∧ e∗i+n ,
i=1
where e∗i ∧ e∗i+n is the alternating 2-form on E defined as follows: if p < q

then (e∗i ∧ e∗i+n )(ep, eq )
= 1 if i = p and i + n = q and 0 otherwise. If L is an
A-linear endomorphism of E, then L∗ ω is by definition equal to ω ◦ (L, L).
Furthermore if L∗ ω = ω, then L is called symplectic. A polynomial map
F := (F1, . . . , F2n) ∈ A2n is called symplectic if the A-linear map on E
defined by the matrix (JF )t is symplectic.
Theorem 2.31. There is equivalence between
i) F is symplectic.
ii) F ∗ : A → A, p → p(F ) is an endomorphism of Pn (R).
Furthermore, if i) or ii) holds and n! is a unit in R, then detJF = 1.
The proof of this result is based on the following lemma. If B is any bilinear
form on E we write M(e) (B) to denote the matrix (B(ei, ej ))1≤i,j≤2n .
Lemma 2.32. Let F = (F1, . . . , F2n ) ∈ A2n . Then
M(e) (((JF )t)∗ ω) = ({Fi, Fj })1≤i,j≤2n .

Proof. Let 1 ≤ p, q ≤ 2n. Then (((J F )t)∗ω)(ep , eq ) = ω((JF )tep, (J F )teq ).
P2n ∂Fp
Observe (JF )tep = j=1 e . So
∂xj j
∂Fp ∂Fq ∂Fp ∂Fq

(e∗i ∧ e∗i+n )((JF )tep, (J F )teq ) = − ,
∂xi ∂xi+n ∂xi+n ∂xi
which gives the desired result.
Corollary 2.33. Let F = (F1, . . . , F2n) ∈ A2n . There is equivalence
between
i) F is symplectic i.e. ((JF )t)∗ ω = ω.
ii) {Fi , Fj } = {xi , xj } for all 1 ≤ i, j ≤ 2n.
iii) F ∗ is an endomorphism of Pn (R).
6. ENDOMORPHISMS OF CANONICAL POISSON ALGEBRAS 41
Proof. By taking F = (x1 , . . . , x2n ) in the above lemma we see that

M(e) (ω) = ({xi , xj })1≤i,j≤2n . Furthermore, since two bilinear forms B1 and
B2 are equal if and only if their matrices M(e) (B1 ) and M(e) (B2 ) are equal,
the equivalence of i) and ii) follows from the lemma above. Finally, using
that F ∗ is a homomorphism with F ∗(xi ) = Fi for all i, the implication
iii)→ii) is obvious and the implication ii)→iii) follows from the fact that
the Poisson bracket is bilinear, antisymmetric and satisfies Leibniz’ rule i.e.
{a, bc} = {a, b}c + {a, c}b for all a, b, c ∈ A.
Proof of Theorem 2.31. The first statement follows from Corollary
2.33. So let F be symplectic and let as before L be the A-linear map of
E = A2n defined by the matrix (J F )t. Put v := e∗1 ∧ . . . ∧ e∗2n the standard
n(n−1)
volume form on E. Since ω n = n!(−1) 2 v (see for example [32], Example
1.4, page 123) and n! is a unit in R, it follows that ω n is a volume form on
E. Furthermore, since F is symplectic L∗ ω = ω and hence
L∗ (ω n ) = (L∗ (ω))n = ω n .
On the other hand, it is well-known that since ω n is a volume form we have
that L∗ (ω n ) = (detL)ω n (see Exercise on page 21 of [32]). Since {ω n } forms
a basis of the A-module of all 2n-forms on E, it follows that det(L) = 1.
Since det(L) = det(JF )t = det(JF ) we get det(J F ) = 1, as desired.
CHAPTER 3
Basics from algebraic geometry
Hanspeter Kraft
1. Affine varieties
Regular functions. Our base field is the field C of complex numbers.
Every polynomial p ∈ C[x1, . . . , xn ] can be regarded as a C-valued function
on Cn in the usual way:
a = (a1, . . . , an ) 7→ p(a) = p(a1, . . . , an ).
These functions will be called regular. More generally, let V be a C-vector
space of dimension dim V = n < ∞.
Definition 3.1. A C-valued function f : V → C is called regular if f is
given by a polynomial p ∈ C[x1, . . . , xn ] with respect to one and hence all
bases of V . This means that for a given basis v1 , . . . , vn of V we have
f (a1 v1 + · · · + an vn ) = p(a1 , . . . , an )
for a suitable polynomial p. The algebra of regular functions on V will be
denoted by O(V ).
By our definition, every choice of a basis (v1, v2, . . . , vn ) of V defines an
∼
isomorphism C[x1, . . . , xn ] → O(V ) by identifying xi with the ith coordinate
function on V defined by the basis, i.e.,
xi(a1 v1 + a2v2 + · · · + an vn ) := ai .
Another way to express this is by remarking that the linear functions on V
are regular and thus the dual space V ∗ := Hom(V, C) is a subspace of O(V ).
So if (v1, v2, . . . , vn ) is a basis of V and (x1, x2 , . . . , xn ) the dual basis of V ∗
then O(V ) = C[x1, x2, . . . , xn ] and the linear functions xi are algebraically
independent.
Example 3.1. Denote by Mn = Mn (C) the complex n × n-matrices so
that O(Mn ) = C[xij | 1 ≤ i, j ≤ n]. Consider det(tEn − X) as a polynomial
43
44 3. BASICS FROM ALGEBRAIC GEOMETRY
in C[t, xij , i, j = 1, . . . , n] where X := (xij ). Developing this as a polynomial

in t we find
det(tEn − X) = tn − s1 tn−1 + s2 tn−2 − · · · + (−1)n sn
with polynomials sk in the variables xij . This defines regular functions sk ∈

O(Mn ) which are homogeneous of degree k. Of course, we have s1 (A) =
tr(A) = a11 + · · · + ann and sn (A) = det(A) for any matrix A ∈ Mn .
The same construction applies to End(V ) for any finite dimensional vector
space V and defines regular function sk ∈ O(End(V )).
Example 3.2. Consider the vector space Pn of unitary polynomials of

degree n:
Pn := {tn − a1 tn−1 + a2tn−2 − · · · + (−1)n an | a1, · · · , an ∈ C} ' Cn .
There is a regular function D ∈ O(Pn ), the discriminant, with the following

property: D(p) = 0 for a p ∈ Pn if and only if p has a multiple root.
Q
Proof. Expanding ni=1 (t − yi ) = tn − σ1(y)tn−1 + · · · + (−1)n σn (y) we
see that the polynomials σj (y) are the elementary symmetric polynomials in
n variables y1, . . . , yn , i.e.
X
σk = σk (y) = σk (y1 , . . . , yn ) := yi1 yi2 · · · yik .
i1 <i2 <···<ik
Q
Define D̃ := i<j (yi −yj )2. Since D̃ is symmetric it can be (uniquely) written
as a polynomial in the elementary symmetric functions (see [6, Chap. 14,
Theorem 3.4]): D̃(y1, . . . , yn ) = D(σ1 , . . . , σn ) with a suitable polynomial D.
By construction, D has the required property.
Example 3.3. We denote by Altn ⊆ Mn the subspace of alternating

matrices:
Altn := {A ∈ Mn | At = −A}.
There is a regular function Pf ∈ O(Alt2m ), the Pfaffian, with the following

property: det(A) = Pf(A)2 for all A ∈ Alt2m. Usually, the sign of
the Pfaffian

J
0 −1
is determined by requiring that Pf( . . . ) = 1 where J := .
J
1 0
1. AFFINE VARIETIES 45
Proof. It is well-known that for any alternating matrix A with entries

in an arbitrary field K there is a g ∈ GLn (K) such that
 
J
 .. 
 . 
t  
(1.1) gAg =  J .
 
 0 
..
.
Now take K = C(xij | 1 ≤ i < j ≤ n = 2m) and put
 
0 x12 x13 · · · x1n
 −x12 0 x23 · · · x2n
 
A :=  −x
 . 13
−x23 0 · · · x3n .
 .. .. .. .. 
. . . 
−x1n −x2n −x3n ··· 0
Then there is a g ∈ GLn (K) such that gAg t has the form given in (1.1). It
follows that the polynomial det(A) ∈ K[xij | 1 ≤ i < j ≤ n] equals det(g)−2 ,
the square of a rational function, and the claim follows.
Exercise 1. For a = (a1, a2, . . . , an ) ∈ Cn denote by eva : O(Cn ) → C the
evaluation map f 7→ f (a). Then the kernel of eva is the maximal ideal
ma := (x1 − a1 , x2 − a2 , . . . , xn − an ).
Exercise 2. Let W ⊆ O(V ) be a finite dimensional subspace. Then the linear
functions evv |W for v ∈ V span the dual space W ∗.
Zero sets and Zariski topology. We now define the basic object of
algebraic geometry, namely the zero set of regular functions. Let V be a finite
dimensional vector space.
Definition 3.2. If f ∈ O(V ) then we define the zero set of f by
V(f ) := {v ∈ V | f (v) = 0} = f −1 (0).
More generally, the zero set of f1 , f2 , . . . , fs ∈ O(V ) or of a subset S ⊆ O(V )
is defined by
\
s
V(f1 , f2 , . . . , fs ) := V(fi ) = {v ∈ V | f1(v) = · · · = fs (v) = 0}
i=1
or
V(S) := {v ∈ V | f (v) = 0 for all f ∈ S}.
Remark 3.3. The following properties of zero sets follow immediately

from the definition.
(1) Let S ⊆ O(V ) and denote by a := (S) ⊆ O(V ) the ideal generated
by S. Then V(S) = V(a).
(2) If S ⊆ T ⊆ O(V ) then V(S) ⊇ V(T ).
(3) For any family (Si )i∈I of subsets Si ⊆ O(V ) we have
\ [
V(Si ) = V( Si ).
i∈I i∈I
Example 3.4. P (1) SLn (C) = V(det −1) ⊆ Mn (C).

(2) On (C) = V( nν=1 xiν xjν − δij | 1 ≤ i ≤ j ≤ n).
(3) If f = f (x, y) ∈ C[x, y] is a non-constant polynomial in 2 variables,
then V(f ) ⊆ C is called a plane curve. In order to visualize a plane
curve, we usually draw a real picture ⊆ R2.
Lemma 3.4. Let V be a finite dimensional vector space and let a, b be
ideals in O(V ) and (ai | i ∈ I) a family of ideals of O(V ).
(1) If
T a ⊆ b then V(a)P ⊇ V(b).
(2) i∈I V(ai ) = V( i∈I ai ).
(3) V(a) ∪ V(b) = V(a ∩ b) = V(a · b).
(4) V(0) = V and V(1) = ∅.
Proof. Properties (1) and (2) follow from Remark 3.3, and property
(4) is easy. So we are left with property (3). Since a ⊇ a ∩ b ⊇ a · b, it
follows from (1) that V(a) ⊆ V(a ∩ b) ⊆ V(a · b). So it remains to show that
V(a · b) ⊆ V(a) ∪ V(b). If v ∈ V does not belong to V(a) ∪ V(b) then there
are functions f ∈ a and h ∈ b such that f (v) 6= 0 6= h(v). Since f · h ∈ a · b
and (f · h)(v) 6= 0 we see that v ∈
/ V(a · b), and the claim follows.
Definition 3.5. The lemma shows that the subsets V(a) where a runs
through the ideals of O(V ) form the closed sets of topology on V which
is called Zariski topology. From now on all topological terms like “open”,
“closed”, “neighborhood”, “continuous”, etc. will refer to the Zariski topol-
ogy.
Example 3.5. (1) The nilpotent cone N ⊆ Mn consisting of all
nilpotent matrices is closed and is a cone, i.e. stable under mul-
tiplication with scalars. E.g. for n = 2 we have
N = V(x11 + x22, x11x22 − x12x21) ⊆ M2.
(r)
(2) The subset Mn ⊆ Mn of matrices of rank ≤ r are closed cones.
(3) The set of polynomials f ∈ Pn with a multiple root is closed (see

Example 3.2).
(4) The closed subsets of C are the finite sets together with C. So the
non-empty open sets of C are the cofinite sets.
Exercise 3. A regular function f ∈ O(V ) is called homogeneous of degree d
if f (tv) = td f (v) for all t ∈ C and all v ∈ V .
(1) A polynomial f ∈ C[x1, . . . , xn] is homogeneous of degree d as a regular
function on Cn if and only if all monomials occurring in f have degree d.
(2) Assume that the ideal a ⊆ O(V ) is generated by homogeneous functions.
Then the zeros set V(a) ⊆ V is a cone.
(3) Conversely, if X ⊆ V is a cone, then the ideal I(X) can be generated by
homogeneous functions.
Exercise 4. Show that the subset A := {(n, m) ∈ C2 | n, m ∈ Z and m ≥
n ≥ 0} is Zariski-dense in C2.
Definition 3.6. Let X ⊆ V be a closed subset. A regular function on
X is defined to be the restriction of a regular function on V :
O(X) := {f |X | f ∈ O(V )}.
The kernel of the (surjective) restriction map res : O(V ) → O(X) is called
the ideal of X:
I(X) := {f ∈ O(V ) | f (x) = 0 for all x ∈ X}.
∼
Thus we have a canonical isomorphism O(V )/I(X) → O(X).
Remark 3.7. Every finite dimensional C-vector space V carries a natural
topology given by the choice of a norm or a hermitian scalar product. We will
call it the C-topology. Since all polynomials are continuous with respect to
the C-topology we see that the C-topology is finer than the Zariski topology.
Exercise 5. Show that every non-empty open set in Cn is dense in the C-
topology. (Hint: Reduce to the case n = 1 where the claim follows from Exam-
ple 3.5(4).)
Remark 3.8. In the Zariski topology the finite sets are closed. This
follows from the fact that for any two different points v, w ∈ V one can find
a regular function f ∈ O(V ) such that f (v) = 0 and f (w) 6= 0. (One says
that the regular functions separate the points.) But the Zariski topology is
not Hausdorff (see the following exercise).
Exercise 6. Let U, U 0 ⊆ Cn be two non-empty open sets. Then U ∩ U 0 is
non-empty, too. In particular, the Zariski topology is not Hausdorff.
Exercise 7. Consider a polynomial f ∈ C[x0 , x1, . . . , xn ] of the form f =

x0 − p(x1 , . . ., xn ), and let X = V(f ) be its zero set. Then I(X) = (f ) and
O(X) ' C[x1 , . . ., xn ].
Hilbert’s Nullstellensatz. The famous Nullstellensatz of Hilbert ap-

pears in many different forms which are all more or less equivalent. We will
give some of them which are suitable for our purposes.
√
Definition 3.9. If a is an ideal of an arbitrary ring R, its radical a is
defined by
√
a := {r ∈ R | rm ∈ a for some m > 0}.
√
The ideal a is perfect if a = a.
√ p√ √
√ It is easy to see that a is an ideal and that a = a. Moreover,
a = R implies that a = R.
Theorem 3.10 (Hilbert’s Nullstellensatz). Let a ⊆ O(V ) be an ideal
and X := V(a) ⊆ V its zero set. Then
√
I(X) = I(V(a)) = a.
A first consequence is that every√strict ideal has a non-empty zero set,
because X = V(a) = ∅ implies that a = I(X) = O(V ) and so a = O(V ).
Corollary 3.11. For every ideal a 6= O(V ) we have V(a) 6= ∅.
Let m ⊆ C[x1, . . . , xn ] be a maximal ideal and a = (a1 , . . . , an ) ∈ V(m)
which exists by the previous corollary. Then m ⊇ (x1 − a1, . . . , xn − an ) and
so these two are equal.
Corollary 3.12. Every maximal ideal m of C[x1, . . . , xn ] is of the form
m = (x1 − a1, . . . , xn − an ).
Another way to express this is by using the evaluation map
evv : O(V ) → C
(see Exercise 1).
Corollary 3.13. Every maximal ideal of O(V ) equals the kernel of the
evaluation map evv : O(V ) → C at a suitable v ∈ V .
Exercise 8. If X ⊆ V is a closed subset and m ⊆ O(X) a maximal ideal then
O(X)/m = C. Moreover, m = ker(evx : f 7→ f (x)) for a suitable x ∈ X.
Proof of Theorem 3.10. We first prove Corollary 3.12 which implies

Corollary 3.13 as we have seen above, and also Corollary 3.11.
Let m ⊆ C[x1, . . . , xn ] be a maximal ideal and denote by
K := C[x1, . . . , xn ]/m its residue class field. Then K contains C and has
a countable C-basis, because C[x1, . . . , xn ] does. If K 6= C and p ∈ K \ C
1
then p is transcendental over C. It follows that the elements ( p−a | a ∈ C)
from K form a non-countable set of linearly independent elements over C.
This contradiction shows that K = C. Thus xi + m = ai + m for a suitable
ai ∈ C (for i = 1, . . . , n), and so m = (x1 − a1, . . . , xn − an ). This proves
Corollary 3.12.
To get the theorem, we use the so-called trick of Rabinowich. Let a ⊆
C[x1, . . . , xn ] be an ideal and assume that the polynomial f vanishes on V(a).
Now consider the polynomial ring R := C[x0, x1, . . . , xn ] in n + 1 variables
and the ideal b := (a, 1 − x0f ) generated by 1 − x0 f and the elements of a.
Clearly, V(b) = ∅ and so 1 ∈ b. This means that we can find an equation of
the form X
hi fi + h(1 − x0 f ) = 1
i
where fi ∈ a and hi , h ∈ R. Now we substitute f1 for x0 and find
X 1
hi ( , x1, . . . , xn )fi = 1.
i
f
P
Clearing denominators finally gives i h̃i fi = f m , i.e., f m ∈ a, and the claim
follows.
Corollary 3.14. For
√ any ideal a ⊆ O(V ) and its zero set X := V(a)
we have O(X) = O(V )/ a.
Exercise 9. Let a ⊆ R be an ideal of a (commutative) ring R. Then a is
perfect if and only if the residue class ring R/a has no nilpotent elements different
from 0.
Example 3.6. Let f ∈ C[x1, . . . , xn ] be an arbitrary polynomial and
consider
p its decomposition into irreducible factors: f = pr11 pr22 · · · prss . Then
(f ) = (p1 p2 · · · ps ) and so the ideal (f ) is perfect if and only if the polyno-
mial f is square-free. In particular, if f ∈ C[x1, . . . , xn ] is irreducible, then
O(V(f )) ' C[x1, . . . , xn ]/(f ). A closed subset of the form V(f ) is called a
hypersurface.
Example 3.7. We have O(SLn (C)) ' O(Mn )/(det −1) because the poly-
nomial det −1 is irreducible. (In fact, if det −1 = f1 · f2 then each factor fi
is linear in the variables which occur. But if xi0 j0 occurs in f1 then all the
variables xij0 and xi0 j have to occur in f1 , too, since det −1 does not contain
products of the form xij0 xi0 j0 and xi0 j xi0 j 0 . This implies that all variables
occur in f1 , hence f2 is a constant.)
Example 3.8. Consider the plane curve C := V(y 2 − x3 ) which is called
∼
Neil’s parabola. Then O(C) ' C[x, y]/(y 2 − x3 ) → C[t2, t3] ⊆ C[t] where the
second isomorphism is given by ρ : x 7→ t3, y 7→ t2.
Proof. Clearly, y 2 − x3 ∈ ker ρ. For any f ∈ C[x, y] we can write f =
f0 (x) + f1 (x)y + h(x, y)(y 2 − x3). If f ∈ ker ρ then 0 = ρ(f ) = f0 (t2) + f1 (t2 )t3
and so f0 = f1 = 0. This shows that ker ρ = (y 2 − x3 ), and the claim
follows.
Exercise 10. Let C ⊆ C2 be the plane curve defined by y − x2 = 0. Then
I(C) = (y − x2 ) and O(C) is a polynomial ring in one variable.
Exercise 11. Let D ⊆ C2 be the zero set of xy − 1. Then O(D) is not
∼
isomorphic to a polynomial ring, but there is an isomorphism O(D) → C[t, t−1 ].
Exercise 12. Consider the “parametric curve”
Y := {(t, t2, t3) ∈ C3 | t ∈ C}.
Then Y is closed in C3 . Find generators for the ideal I(Y ) and show that O(Y ) is
isomorphic to the polynomial ring in one variable.
Another important consequence of the “Nullstellensatz” is an one-to-one
correspondence between closed subsets of Cn and perfect ideals of the coor-
dinate ring C[x1, . . . , xn ].
Corollary 3.15. The map X 7→ I(X) defines an inclusion-reversing
bijection
∼
{X ⊆ V closed} → {a ⊆ O(V ) perfect ideal}
whose inverse map is given by a 7→ V(a). Moreover, for any finitely generated
reduced C-algebra R there is a closed subset X ⊆ Cn for some n such that
O(X) is isomorphic to R.
√
Proof. The first part is clear since V(I(X)) = X and I(V(a)) = a for
any closed subset X ⊆ V and any ideal a ⊆ O(V ).
If R is a reduced and finitely generated C-Algebra, R = C[f1, . . . , fn ],
then R ' C[x1, x2 , . . . , xn ]/a where a is the kernel
√ of the homomorphism
defined by xi 7→ fi . Since R is reduced we have a = a and so O(V(a)) '
C[x1, . . . , xn ]/a ' R.
Exercise 13. Let X ⊆ V be a closed subset and f ∈ O(X) a regular function

such that f (x) 6= 0 for all x ∈ X. Then f is invertible in O(X), i.e. the C-valued
function x 7→ f (x)−1 is regular on X.
Exercise 14. Every closed subset X ⊆ Cn is quasi-compact, i.e., every cov-
ering of X by open sets contains a finite covering.
Exercise 15. Let X ⊆ Cn be a closed subset. Assume that there is no non-
constant invertible regular function on X. Then every non-constant f ∈ O(X)
attains all values in C, i.e. f : X → C is surjective.
Affine varieties. We have seen in the previous section that every closed
subset X ⊆ V (or X ⊆ Cn ) is equipped with an algebra of C-valued functions,
namely the coordinate ring O(X). We first remark that O(X) determines the
topology of X. In fact, define for every ideal a ⊆ O(X) the zero set in X by
VX (a) := {x ∈ X | f (x) = 0 for all f ∈ a}.
Clearly, we have VX (a) = V(ã) ∩ X where ã ⊆ O(V ) is an ideal which maps
surjectively onto a under the restriction map. This shows that the sets VX (a)
are the closed sets of the topology on X induced by the Zariski topology of
V . Moreover, the coordinate ring O(X) also determines the points of X since
they are in one-to-one correspondence with the maximal ideals of O(X):
x ∈ X 7→ mx := ker evx ⊆ O(X)
where evx : O(X) → C is the evaluation map f 7→ f (x). This leads to the
following definition of an “abstract” zero set, not embedded in a vector space.
Definition 3.16. A set Z together with a C-algebra O(Z) of C-valued
functions on Z is called an affine variety if there is a closed subset X ⊆ Cn
∼
for some n and a bijection ϕ : Z → X which identifies O(X) with O(Z), i.e.,
ϕ∗ : O(X) → O(Z) given by f 7→ f ◦ ϕ, is an isomorphism.
The functions from O(Z) are called regular, and the algebra O(Z) is called
coordinate ring of Z or algebra of regular functions on Z.
The affine variety Z has a natural topology, the Zariski topology, the
closed sets being the zero sets
VZ (a) := {z ∈ Z | f (z) = 0 for all f ∈ a}
where a runs through the ideals of O(Z). It follows from what we said above
∼
that the bijection ρ : Z → X is a homeomorphism with respect to the Zariski
topology.
Clearly, every closed subset X ⊆ V or X ⊆ Cn together with its co-
ordinate ring O(X) is an affine variety. More generally, if X is an affine
variety and Y ⊆ X a closed subset, then Y together with the restrictions

O(Y ) := {f |Y | f ∈ O(X)} is an affine variety, called a closed subvariety.
Less trivial examples are the following.
Example 3.9. Let M be a finite set and define O(M) := Maps(M, C)
to be the set of all C-valued functions on M. Then (M, O(M)) is an affine
variety and O(M) is isomorphic to a product of copies of C. This follows
immediately from the fact that any finite subset N ⊆ Cn is closed and that
O(N ) = Maps(N, C).
Example 3.10. Let X be a set and define the symmetric product Symn (X)
to be the set of unordered n-tuples of elements from X, i.e.,
Symn (X) = X × X × · · · × X/ ∼
where (a1 , a2, . . . , an ) ∼ (b1 , b2, . . . , bn ) if and only if one is a permutation of
the other.
In case X = C we define O(Symn (C)) to be the symmetric polynomials
in n variables and claim that Symn (C) is an affine variety.
To see this consider the map
Φ : Cn → Cn , a = (a1, . . . , an ) 7→ (σ1(a), σ2(a), . . . , σn (a))
where σ1, . . . , σn are the elementary symmetric polynomials (see Example 3.2).
It is easy to see that Φ is surjective and that Φ(a) = Φ(b) if and only if a ∼ b.
∼
Thus, Φ defines a bijection ϕ : Symn (C) → Cn , and the pull-back of the reg-
∼
ular functions on Cn are the symmetric polynomials: ϕ∗ : C[x1, . . . , xn ] →
O(Symn (C)).
Exercise 16. Let Z be an affine variety with coordinate ring O(Z). Then
every polynomial f ∈ O(Z)[t] with coefficients in O(Z) defines a function on the
product Z × C in the usual way:
m
X m
X
k
f= fk t : (z, a) 7→ fk (z)ak ∈ C.
k=0 k=0
Show that Z × C together with O(Z)[t] is an affine variety.

(Hint: For any ideal a ⊆ C[x1, . . . , xn ] there is a canonical isomorphism
∼
C[x1, . . . , xn , t]/(a) → (C[x1, . . . , xn ]/a)[t].)
Exercise 17. For any affine variety Z there is an inclusion-reversing bijection
∼
{A ⊆ Z closed} → {a ⊆ O(Z) perfect ideal}
given by A 7→ I(A) := {f ∈ O(Z) | f |A = 0} (cf. Corollary 3.15).
For the last example we start with a reduced and finitely generated C-
algebra R. Denote by Spec R the set of maximal ideals of R:
Spec R := {m | m ⊆ R a maximal ideal}.
We know from the “Nullstellensatz” (see Exercise 8) that R/m = C for all
maximal ideals m ∈ Spec R. This allows to identify the elements from R with
C-valued functions on Spec R: For f ∈ R and m ∈ Spec R we define
f (m) := f + m ∈ R/m = C.
Proposition 3.17. Let R be a reduced and finitely generated C-algebra.
Then the set of maximal ideals Spec R together with the algebra R considered
as functions on Spec R is an affine variety.
Proof. We have already seen earlier that every such algebra R is iso-
morphic to the coordinate ring of a closed subset X ⊆ Cn . The claim then
∼
follows by using the bijection X → Spec O(X), x 7→ mx = ker evx , and re-
marking that for f ∈ O(X) and x ∈ X we have f (x) = evx (f ) = f + mx , by
definition.
Exercise 18. Denote by Cn the n-tuples of complex numbers up to sign, i.e.,
Cn := Cn / ∼ where (a1, . . . , an) ∼ (b1, . . . , bn) if ai = ±bi for all i. Then every
polynomial in C[x21, x22, . . ., x2n ] is a well-defined function on Cn . Show that Cn
together with these functions is an affine variety.
(Hint: Consider the map Φ : Cn → Cn , (a1, . . . , an ) 7→ (a21, . . . , a2n ) and proceed
like in Example 3.10.)
Although every affine variety is “isomorphic” to a closed subset of Cn
for a suitable n, there are many advantages to look at these objects and
not only at closed subsets. In fact, an affine variety can be identified with
many different closed subsets of this form (see the following Exercise 19),
and depending on the questions we are asking, one of them might be more
useful than another. We will even see in the following section that certain
open subsets are affine varieties in a natural way.
On the other hand, whenever we want to prove some statements for an
affine variety X we can always assume that X = V(a) ⊆ Cn so that the
regular functions on X appear as restrictions of polynomial functions. This
will be helpful in the future and quite often simplify the arguments.
Exercise 19. Let X be an affine variety. Show that every choice of a generating
system f1 , f2, . . . , fn ∈ O(X) of the algebra O(X) consisting of n elements defines
an identification of X with a closed subset V(a) ⊆ Cn .
(Hint: Consider the map X → Cn given by x 7→ (f1(x), f2(x), . . ., fn (x)).)
Special open sets. Let X be an affine variety and f ∈ O(X). Define

the open set Xf ⊆ X by
Xf := X \ VX (f ) = {x ∈ X | f (x) 6= 0}.
An open set of this form is called a special open set.
Lemma 3.18. The special open sets of an affine variety X form a basis
of the topology.
Proof. If U ⊆ X is open and x ∈ U , then X \ U is closed and does not
contain x. Thus, there is a regular function f ∈ O(X) vanishing on X \ U
such that f (x) 6= 0. This implies x ∈ Xf ⊆ U .
Given a special open set Xf ⊆ X we see that f (x) 6= 0 for all x ∈ Xf
and so the function f1 is well-defined on Xf .
Proposition 3.19. Denote by O(Xf ) the algebra of functions on Xf

generated by f1 and the restrictions h|Xf of regular functions h on X:
1 1
O(Xf ) := C[ , {h|Xf | h ∈ O(X)}] = O(X)|Xf [ ].
f f
∼
Then (Xf , O(Xf )) is an affine variety and O(Xf ) → O(X)[t]/(f · t − 1).
Proof. Let X = V(a) ⊆ Cn and define
X̃ := V(a, f · xn+1 − 1) ⊆ Cn+1 .
It is easy to see that the projection pr : Cn+1 → Cn onto the first n coordi-
∼ ∼
nates induces a bijective map X̃ → Xf whose inverse ϕ : Xf → X̃ is given
by
ϕ(x1, . . . , xn ) = (x1 , . . . , xn , f(x1 , . . . , xn )−1 ).
The following commutative diagram now shows that ϕ∗ (O(X̃)) is generated
by ϕ∗(xn+1 ) = f1 and the restrictions h|Xf (h ∈ O(X)).
⊆
X̃ X̃ −−−→ Cn+1
x  closed 

ϕ'
 pr
y y
⊆ ⊆
Xf −−−→ X −−−→ Cn
open closed
This proves the first claim. For the second, we have to show that the canonical
homomorphism O(X)[t]/(f ·t−1) → O(X̃ ) is an isomorphism. In other words,
P
every function h = m i
i=0 hi t ∈ O(X)[t] which vanishes
P on X̃ is divisible by
m−i
f · t − 1. Since f |X̃ is invertible we first obtain i hi f = 0 which implies
X
m X
m−1
m m−i
h =h−t hi f = hi ti (1 − f m−i tm−i ),
i=0 i=0
and the claim follows.

Example 3.11. The group GLn (C) is a special open set of Mn (C), hence
GLn (C) is an affine variety with coordinate ring O(GLn (C)) = C[{xij | 1 ≤
1
i, j ≤ n}, det ]. In particular, C∗ := GL1 = C \ {0} is an affine variety with
coordinate ring C[x, x−1].
Exercise 20. Let R be an arbitrary C-algebra. For any element s ∈ R define
Rs := R[x]/(s · x − 1).
(1) Describe the kernel of the canonical homomorphism ι : R → Rs .
(2) Prove the universal property: For any homomorphism ρ : R → A such
that ρ(s) is invertible in A there is a unique homomorphism ρ̄: Rs → A
such that ρ̄ ◦ ι = ρ.
(3) What happens if s is a zero divisor and what if s is invertible?
Decomposition into irreducible components. We start with a purely

topological notion.
Definition 3.20. A topological space T is called irreducible if it cannot
be decomposed in the form T = A ∪ B where A, B $ T are strict closed
subsets. Equivalently, every non-empty open subset is dense.
Lemma 3.21. Let X ⊆ Cn be a closed subset. Then the following are
equivalent:
(i) X is irreducible.
(ii) I(X) is a prime ideal.
(iii) O(X) is a domain, i.e., has no zero-divisor.
Proof. (i)⇒(ii): If I(X) is not prime we can find two polynomials f, h ∈
C[x1, . . . , xn ] \ I(X) such that f · h ∈ I(X). This implies that
X ⊆ V(f · h) = V(f ) ∪ V(h),
but X is neither contained in V(f ) nor in V(h). Thus
X = (V(g) ∩ X) ∪ (V(h) ∩ X)
is a decomposition into strict closed subsets, contradicting the assumption.
(ii)⇒(iii): This is clear since O(X) = C[x1, . . . , xn ]/I(X).
(iii)⇒(i): If X = A ∪ B is a decomposition into strict closed subsets there

are non-zero functions f, h ∈ O(X) such that f |A = 0 and h|B = 0. But then
f · h vanishes on all of X and so f · h = 0. This contradicts the assumption
that O(X) does not contain zero-divisor.
Example 3.12. Let f ∈ C[x1, . . . , xn ]. Then the hypersurface V(f ) is
irreducible if and only if f is a power of an irreducible polynomial. This
follows from Example 3.6 and the fact that the ideal (f ) is prime if and
only if f is irreducible. More generally, if f = pr11 pr22 · · · prss is the primary
decomposition, then
V(f ) = V(p1 ) ∪ V(p2 ) ∪ · · · ∪ V(pn )
where each V(pi ) is irreducible, and this decomposition is irredundant, i.e.,
no term can be dropped.
Theorem 3.22. Every affine variety X is a finite union of irreducible
closed subsets Xi :
(1.2) X = X1 ∪ X2 ∪ · · · ∪ Xs .
If this decomposition is irredundant, then the Xi ’s are the maximal irreducible
subsets of X and are therefore uniquely determined.
The unique irredundant decomposition of an affine variety X in the form
(1.2) is called irreducible decomposition and the Xi ’s are called the irreducible
components.
For the proof of the theorem above we first recall that a C-algebra R is
called Noetherian if the following equivalent conditions hold:
(i) Every ideal of R is finitely generated.
(ii) Every strictly ascending chain of ideals of R terminates.
(iii) Every non-empty set of ideals of R contains maximal elements.
(The easy proofs are left to the reader; for the equivalence of (ii) and (iii)
one has to use Zorn’s Lemma.)
The famous “Basissatz” of Hilbert implies that every finitely generated
C-algebra is Noetherian (see [6, Chap. 12, Theorem 5.18]). In particular,
this holds for the coordinate ring O(X) of any affine variety X. Using the
inclusion reversing bijection between closed subsets of X and perfect ideals
of O(X) (see Corollary 3.15 and Exercise 17) we get the following result.
Proposition 3.23. Let X be an affine variety. Then
(1) Every closed subset A ⊆ X is of the form VX (f1 , f2, . . . , fr ).
(2) Every strictly descending chain of closed subsets of X terminates.
(3) Every non-empty set of closed subsets of X contains minimal ele-

ments.
Remark 3.24. It is easy to see that for an arbitrary topological space
T the properties (2) and (3) from the previous proposition are equivalent. If
they hold then T is called Noetherian.
Proof of Theorem 3.22. We first show that such a decomposition
exists. Consider the following set
M := {A ⊆ X | A closed and not a finite union of irreducible closed subsets}.
If M 6= ∅ then it contains a minimal element A0. Since A0 is not irreducible,
we can find strict closed subsets B, B 0 ( A0 such that A0 = B ∪ B 0. But
then B, B 0 ∈
/ M and so both are finite unions of irreducible closed subsets.
Hence A0 is a finite union of irreducible closed subsets, too, contradicting the
assumption.
To show the uniqueness let X = X1 ∪ X2 ∪ · · · ∪ Xs where all Xi are
irreducible closed subsets and assume that the decomposition is irredundant.
Then, clearly, every Xi is maximal. Let Y ⊆ X be a maximal irreducible
closed subset. Then Y = (Y ∩ X1 ) ∪ (Y ∩ X2 ) ∪ · · · ∪ (Y ∩ Xs ) and so
Y = Y ∩ Xj for some j. It follows that Y ⊆ Xj and so Y = Xj because of
maximality.
Remark 3.25. The algebraic counterpart to the decomposition into irre-
ducible components is the following statement about radical ideals in finitely
generated algebras R: Every radical ideal a ⊆ R is a finite intersection of
prime ideals:
a = p1 ∩ p2 ∩ · · · ∩ ps .
If this intersection is irredundant then the pi ’s are the minimal prime ideals
containing a. (The easy proof is left to the reader.)
Example 3.13. Consider the two hypersurfaces H1 := V(xy − z), H2 :=
V(xz − y 2) in C3 and their intersection X := H1 ∩ H2 . Then
X = V(y, z) ∪ C where C := {(t, t2, t3) | t ∈ C},
and this is the irreducible decomposition.
In fact, it is obvious that the x-axis V(y, z) is closed and irreducible
and belongs to X, and the same holds for the curve C (see Exercise 12). If
(a, b, c) ∈ X \ V(y, z) then either b or c is 6= 0. But then b 6= 0 because ab = c.
Hence a = cb−1 and so b2 = ac = c2 b−1 which implies that c2 = b3. Thus
b = (cb−1 )2 and c = (cb−1)3 , i.e. (a, b, c) ∈ C.
Another way to see this is by looking at the coordinate ring:

∼
C[x, y, z]/(xy − z, xz − y 2 ) → C[x, y]/(x2y − y 2).
Now (x2y − y 2) = y(x − y 2) = (y) ∩ (x − y 2) and the ideals (y) and (x − y 2)
are obviously prime with residue class ring isomorphic to a polynomial ring
in one variable. This shows again that X has two irreducible components,
both with coordinate ring isomorphic to C[t].
Exercise 21. The closed subvariety X := V(x2 − yz, xz − x) ⊆ C3 has three
irreducible components. Describe the corresponding prime ideals in C[x, y, z].
Example 3.14. The group O2 := {A ∈ M2 | AAt = E} has two irre-

0 1
ducible components, namely SO2 := O2 ∩ SL2 and · SO2 , and the two
1 0
components are disjoint.
t a b a
In fact, the condition AA = E for A = implies that =
c d b

d a b a b
± . Since det = a2 + b2 we see that SO2 = { | a2 + b2 =
−c −b a −b a

0 1 a b
1} is irreducible as well as · SO2 = { | a2 + b2 = 1} and the
1 0 b −a
claim follows immediately.
Exercise 22. Let G ⊆ GLn be a closed subgroup.
(1) The irreducible components of G are disjoint.
(2) The irreducible component of G containing E is a normal subgroup G0.
(3) The irreducible components are the cosets of G0.
S
Exercise 23. Let X = Xi be the decomposition
S into irreducible components
and let Ui ⊆ Xi be open subsets. Then U := Ui is open in X. It is dense in X
if and only if all Ui are non-empty.
Rational functions and local rings. If X is an irreducible affine vari-

ety then O(X) is a domain by Lemma 3.21. Therefore, we can form the field
of fractions of O(X) which is called the field of rational functions on X and
will be denoted by C(X). Clearly, if X = Cn then C(X) = C(x1 , x2, . . . , xn ),
the rational function field. An irreducible affine variety X is called rational if
its field of rational functions C(X) is isomorphic to a rational function field.
A rational function f ∈ C(X) can be regarded as a function “defined
almost everywhere” on X. In fact, we say that f is defined in x ∈ X if there
are p, q ∈ O(X) such that f = pq and q(x) 6= 0.
Exercise 24. If f ∈ C(C2) = C(x, y) is defined in C2 \ {(0, 0)} then f is

regular.
Exercise 25. Let f be a rational function on the irreducible affine variety X
and denote by Def(f ) ⊆ X the set of points where f is defined.
(1) The set Def(f ) is open in X.
(2) If Def(f ) = X then f is regular on X. (Hint: Look at the “ideal of
denominators” a := {p ∈ O(X) | p · f ∈ O(X)}.)
Example 3.15. Consider the irreducible plane curve C := V(y 2 − x3) ⊆

C and put x̄ := x|C and ȳ := y|C . Then the rational function f := ȳx̄ ∈ C(C)
2
is not defined in (0, 0). The interesting point here is that f has a continuous
extension to all of C with value 0 at (0, 0), even in the C-topology.
∼
Proof. There is an isomorphism O(C) → C[t2, t3 ] (see Example 3.8)
which maps x̄ to t2 and ȳ to t3, and so f = x̄ȳ is mapped to t. Since t ∈
/ C[t2, t3]
the first claim follows from Exercise 25(2) above. The second part is easy
because the map C → C : t 7→ (t2, t3) is a homeomorphism even in the C-
topology.
Assume that X is irreducible and let x ∈ X. Define
OX,x := {f ∈ C(X) | f is defined in x}.
It is easy to see that OX,x is the localization of O(X) at the maximal ideal mx .
(For the definition of the localization of a ring at a prime ideal and, more
generally, for the construction of rings of fractions we refer to [26, I.2.1].)
This example motivates the following definition.
Definition 3.26. Let X be an affine variety and x ∈ X an arbitrary
point. Then the localization O(X)mx of the coordinate ring O(X) at the
maximal ideal in x is called the local ring of X at x. It will be denoted by
OX,x , its unique maximal ideal by mX,x .
We will see later that the local ring of X at x completely determines X
in a neighborhood of x.
T
Exercise 26. If X is irreducible, then O(X) = x∈X OX,x .
Exercise 27. Let X be an affine variety, x ∈ X a point and X 0 ⊆ X the
union of irreducible components of X passing through x. Then the restriction map
∼
induces a natural isomorphism OX,x → OX 0 ,x .
Exercise 28. Let R be an algebra and µ : R → RS the canonical map r 7→ 1r
where RS is the localization at a multiplicatively closed subset 1 ∈ S ⊆ R (0 ∈
/ S).
(1) If a ⊆ R is an ideal and aS := RS µ(a) ⊆ RS then
µ−1 (µ(a)) = µ−1 (aS ) = {b ∈ R | sb ∈ a for some s ∈ S}.

∼
Moreover, (R/a)S̄ → RS /aS where S̄ is the image of S in R/a.
(Hint: For the second assertion use the universal property of the localiza-
tion.)
(2) If m ⊆ R is a maximal ideal and S := R \ m, then mS is the maximal
∼
ideal of RS and the natural maps R/mk → RS /mkS are isomorphisms for
all k ≥ 1.
(Hint: The image S̄ in R/mk consists of invertible elements.)
Exercise 29. Let p < q be positive integers with no common divisor and
define Cp,q := {(tp , tq ) | t ∈ C} ⊆ C2 . Then Cp,q is a closed irreducible plane curve
which is rational, i.e. C(Cp,q ) ' C(t). Moreover, O(Cp,q ) is a polynomial ring if
and only if p = 1.
Exercise 30. Consider the elliptic curve E := V(y 2 − x(x2 − 1)) ⊆ C2 . Show
that E is not rational, i.e. that C(E) is not isomorphic to C(t). (Hint: If C(E) =
C(t) then there are rational functions f (t), h(t) which satisfy the equation f (t)2 =
h(t)(h(t)2 − 1).)
2. Morphisms
Morphisms and comorphisms. In the previous sections we have de-
fined and discussed the main objects of algebraic geometry, the affine vari-
eties. Now we have to introduce the “regular maps” between affine varieties
which should be compatible with the concept of regular functions.
Definition 3.27. Let X, Y be affine varieties. A map ϕ : X → Y is called
regular or a morphism if the pull-back of a regular function on Y is regular
on X:
f ◦ ϕ ∈ O(X) for all f ∈ O(Y ).
Thus we obtain a homomorphism ϕ∗ : O(Y ) → O(X) of C-algebras given by
ϕ∗ (f ) := f ◦ ϕ, which is called comorphism of ϕ.
A morphism ϕ is called an isomorphism if ϕ is bijective and the inverse
map ϕ−1 is also a morphism. If, in addition, Y = X then ϕ is called an
automorphism.
Exercise 31. Let g ∈ GLn be an invertible matrix. Then left multiplication
A 7→ gA, right multiplication A 7→ Ag and conjugation A 7→ gAg −1 are automor-
phisms of Mn .
2. MORPHISMS 61
Example 3.16. A map ϕ = (f1 , f2 , . . . , fm ) : Cn → Cm is regular if and

only if the components fi are polynomials in C[x1, . . . , xn ]. This is clear, since
ϕ∗ (yj ) = fj where y1, y2 , . . . , ym are the coordinate functions on Cm .
More generally, let X be an affine variety and ϕ = (f1 , . . . , fm ) : X → Cm
a map. Then ϕ is a morphism if and only if the components fj are regular
functions on X. (This is clear since fj = ϕ∗ (yj ).)
If a morphism ϕ = (f1 , f2 , . . . , fm ) : Cn → Cm maps a closed subset
X ⊆ Cn into a closed subset Y ⊆ Cm then the induced map ϕ̄ : X → Y is
clearly a morphism, just by definition. This holds in a slightly more general
setting, as claimed in the next exercise.
Exercise 32. Let ϕ : X → Y be a morphism. If X 0 ⊆ X and Y 0 ⊆ Y are closed
subvarieties such that ϕ(X 0) ⊆ Y 0 then the induced map ϕ0 : X 0 → Y 0 , x 7→ ϕ(x),
is again a morphism. The same holds if X 0 and Y 0 are special open sets.
These examples have the following converse which will be useful in many
applications.
Lemma 3.28. Let X ⊆ Cn and Y ⊆ Cm be closed subvarieties and
let ϕ : X → Y be a morphism. Then there are polynomials f1 , . . . , fm ∈
C[x1, . . . , xn ] such that the following diagram commutes:
Φ:=(f1 ,...,fm )
Cn −−−−−−−−→ Cm
x x
⊆ ⊆
 
ϕ
X −−−→ Y
Proof. Let y1 , . . . , ym denote the coordinate functions on Cm . Put ȳj :=
yj |Y and define f¯j := ϕ∗(ȳj ) ∈ O(X), j = 1, . . . , m. Since O(X) =
C[x1, . . . , xn ]/I(X) we can find representatives fj ∈ C[x1, . . . , xn ], i.e. f¯j =
fj +I(X). We claim that the morphism Φ := (f1 , . . . , fm ) : Cn → Cm satisfies
the requirements of the lemma. In fact, let a ∈ X ⊆ Cn and set ϕ(a) =: b =
(b1, . . . , bm ). Then
bj = yj (b) = ȳj (b) = ȳj (ϕ(a)) = ϕ∗ (ȳj )(a) = f¯j (a) = fj (a),
and so ϕ(a) = Φ(a).
Example 3.17. The morphism t 7→ (t2 , t3) from C → C2 induces a
bijective morphism C → C := V(y 2 − x3) which is not an isomorphism (see
Example 3.8).
Similarly there is a morphism ψ : C → D := V(y 2 − x2 − x3) given by
t 7→ (t2 − 1, t(t2 − 1)). This time ψ is surjective, but not injective since
ψ(1) = ψ(−1) = (0, 0).
Exercise 33.
(1) Every morphism C → C∗ is constant.
(2) Describe all morphisms C∗ → C∗ .
(3) Every non-constant morphism C → C is surjective.
(4) An injective morphism C → C is an isomorphism, and the same holds for
morphisms C∗ → C∗.
Exercise 34. The graph of a morphism. Let ϕ : Cn → Cm be a morphism
and define
Γϕ := {(a, ϕ(a)) ∈ Cn+m }.
which is called the graph of the morphism ϕ. Show that Γϕ is closed in Cn+m , that
∼
the projection prCn : Cn+m → Cn induces an isomorphism p : Γϕ → Cn and that
ϕ = prCm ◦p .−1
Proposition 3.29. Let X, Y be affine varieties. The map ϕ 7→ ϕ∗ in-

duces a bijection
∼
Mor(X, Y ) → AlgC (O(Y ), O(X))
between the morphisms from X to Y and the algebra homomorphism from
O(Y ) to O(X).
Remark 3.30. The mathematical term used in the situation above is that
of a contravariant functor from the category of affine varieties and morphisms
to the category of finitely generated reduced C-algebras and homomorphism,
given by X 7→ O(X) and ϕ 7→ ϕ∗ . In particular, we have ϕ∗ (IdX ) = IdO(X)
and (ϕ ◦ ψ)∗ = ψ ∗ ◦ ϕ∗ whenever the expressions make sense. The proposition
above then says that this functor is an equivalence, the inverse functor being
R 7→ Spec R defined in Proposition 3.17. It will be helpful to keep this
“functorial point of view” in mind although it will not play an important
role in the following.
Proof. (a) If ϕ∗1 = ϕ∗2 then, for all f ∈ O(Y ) and all x ∈ X, we get
f (ϕ1 (x)) = ϕ∗1 (f )(x) = ϕ∗2(f )(x) = f (ϕ2 (x)).
Hence, ϕ1 (x) = ϕ2(x) since the regular functions separate the points (Re-
mark 3.8).
(b) Let ρ : O(Y ) → O(X) be an algebra homomorphism. We want to
construct a morphism ϕ : X → Y such that ϕ∗ = ρ. For this we can assume
that Y ⊆ Cm is a closed subvariety. Let ȳj := yj |Y be the restrictions of
the coordinate functions on Cm and define fj := ρ(ȳj ) ∈ O(X). Then we
get a morphism Φ := (f1 , . . . , fm ) : X → Cm such that Φ∗ (yj ) = fj (see
Example 3.16). If h = h(y1, . . . , ym ) ∈ I(Y ) then
h(f1 , . . . , fm ) = h(ρ(ȳ1 ), . . . , ρ(ȳm )) = ρ(h(ȳ1 , . . . , ȳm )) = 0
2. MORPHISMS 63
because h(ȳ1 , . . . , ȳm ) = h|Y = 0 by assumption. Therefore h(Φ(a)) = 0 for

all a ∈ X and all h ∈ I(Y ) and so Φ(X) ⊆ Y . This shows that Φ induces
a morphism ϕ : X → Y such that ϕ∗ (ȳj ) = Φ∗ (yj ) = fj = ρ(ȳj ), and so
ϕ∗ = ρ.
Example 3.18. Let X be an affine variety, V a finite dimensional vector
space and ϕ : X → V a morphism. The linear functions on V form a sub-
space V ∗ ⊆ O(V ) which generates O(V ). Therefore, the induced linear map
ϕ∗ |V ∗ : V ∗ → O(X) completely determines ϕ∗ , and we get a bijection
∼
Mor(X, V ) → Hom(V ∗, O(X)) ϕ 7→ ϕ∗ |V ∗ .
The second assertion follows from Proposition 3.29 and the well-known
“Substitution Principle” for polynomials rings (see [6, Chap. 10, Proposi-
tion 3.4]).
Exercise 35. Show that for an affine variety X the morphisms X → C∗
correspond bijectively to the invertible functions on X.
Exercise 36. Let X, Y be affine varieties and ϕ : X → Y , ψ : Y → X mor-
∼
phisms such that ψ ◦ ϕ = IdX . Then ϕ(X) ⊆ Y is closed and ϕ : X → ϕ(X) is an
isomorphism.
Images, pre-images and fibers. It is easy to see that morphisms are
continuous. In fact, the Zariski topology is the finest topology such that
regular functions are continuous, and since morphisms are defined by the
condition that the pull-back of a regular function is again regular, it imme-
diately follows that morphisms are continuous. We will get this result again
from the next proposition where we describe images and preimages of closed
subsets under morphisms.
Proposition 3.31. Let ϕ : X → Y be a morphism of affine varieties.
(1) If B := VY (S) ⊆ Y is the closed subset defined by S ⊆ O(Y ) then
ϕ−1 (B) = VX (ϕ∗ (S)). In particular, ϕ is continuous.
(2) Let A := V(a) ⊆ X be the closed subset defined by the ideal a ⊆
O(X). Then the closure of the image ϕ(A) is defined by ϕ∗−1 (a) ⊆
O(Y ):
ϕ(A) = VY (ϕ∗−1 (a)).
Proof. For x ∈ X we have
x ∈ ϕ−1 (B) ⇔ ϕ(x) ∈ B ⇔ f (ϕ(x)) = 0 for all f ∈ S,
and this is equivalent to ϕ∗ (f )(x) = 0 for all f ∈ S, hence to x ∈ VX (ϕ∗ (S)),
proving the first claim.
For the second claim, let f ∈ O(Y ). Then

√
f |ϕ(A) = 0 ⇔ f |ϕ(A) = 0 ⇔ ϕ∗(f )|A = 0 ⇔ ϕ∗ (f ) ∈ I(A) = a
The latter is equivalent to the condition that a power of f belongs to ϕ∗−1 (a).
Thus the zero set of ϕ∗−1 (a) equals the closed set ϕ(A).
Exercise 37. If ϕ1, ϕ2 : X → Y are two morphisms, then the “kernel of coin-
cidence”
ker(ϕ1, ϕ2) := {x ∈ X | ϕ1(x) = ϕ2 (x)} ⊆ X
is closed in X
Exercise 38. Let ϕ : X → Y be a morphism of affine varieties.
(1) If X is irreducible, then ϕ(X) is irreducible.
(2) Every irreducible component of X is mapped into an irreducible compo-
nent of Y .
∼
Exercise 39. Let ϕ : X → X be an automorphism and Y ⊆ X a closed subset
such that ϕ(Y ) ⊆ Y . Then ϕ(Y ) = Y and ϕ|Y : Y → Y is an automorphism, too.
A special case of pre-images are the fibers of a morphism ϕ : X → Y . Let
y ∈ Y . Then
ϕ−1 (y) := {x ∈ X | ϕ(x) = y}
is called the fiber of y ∈ Y . By the proposition above, the fiber of y is a closed
subvariety of X defined by ϕ∗ (my ):
ϕ−1 (y) = VX (ϕ∗ (my )).
Of course, the fiber of a point y ∈ Y can be empty. In algebraic terms this
means that ϕ∗ (my ) generates the unit ideal (1) = O(X).
Exercise 40. Let ϕ : X → Y be a morphism of affine varieties. If U ⊆ Y is a
special open set then so is ϕ−1 (U ).
Exercise 41. Describe the fibers of the morphism ϕ : M2 → M2 , A 7→ A2.
(Hint: Use the fact that ϕ(gAg −1) = gϕ(A)g −1 for g ∈ GL2 .)
Definition 3.32. Let ϕ : X → Y be a morphism of affine varieties and
consider the fiber F := ϕ−1 (y) of a point y ∈ ϕ(X) ⊆ Y . Then the fiber F is
called reduced if ϕ∗ (my ) generates a perfect ideal in O(X), i.e. if
q
O(X) · ϕ∗ (my ) = O(X) · ϕ∗ (my ).
The fiber F is called reduced in the point x ∈ F if this holds in the local ring
OX,x , i.e. q
OX,x · ϕ∗ (my ) = OX,x · ϕ∗ (my ).
2. MORPHISMS 65
Example 3.19. Look again at the morphism ϕ : C → C := V(y 2 − x3) ⊆

∼
C2 , t 7→ (t2, t3). Then ϕ∗ is the injection O(C) → C[t2, t3 ] ,→ C[t] and so
p
C[t] · ϕ∗ (m(0,0)) = (t2, t3) ( (t2, t3) = (t).
Thus the zero fiber ϕ−1 (0) is not reduced. On the other hand, all other fibers
are reduced since ϕ induces an isomorphism of C∗ with the special open set
C \ {(0, 0)}(= Cx̄ = Cȳ ), where the inverse map is given by (a, b) 7→ ab .
Exercise 42. Show that all fibers of the morphism ψ : C → D := V(y 2 − x2 −
x ) ⊆ C2, t 7→ (t2 − 1, t(t2 − 1)), are reduced and that ψ induces an isomorphism
3
∼
C \ {1, −1} → D \ {(0, 0)}.
Exercise 43. Consider the following morphism ϕ : SL2 → C3 ,

a b
ϕ( ) := (ab, ad, cd).
c d
(1) The image of ϕ is a closed hypersurface H ⊆ C3 .
(2) The fibers of ϕ are the left cosets of the subgroup

t
T := { | t ∈ C∗}.
t−1
(3) All fibers are reduced.
(Hint: Show that the left multiplication with some g ∈ SL2 induces an automor-
∼
phism λg of H and isomorphisms ϕ−1 (y) → ϕ−1 (λg (y)) for all y ∈ H. This implies
that it suffices to study just one fiber, e.g. ϕ−1 (ϕ(E)).)
Exercise 44. Consider the morphism ϕ : C2 → C2 given by ϕ(x, y) := (x, xy).
(1) ϕ(C2) = C2 \ {(0, y) | y 6= 0} which is not locally closed.
(2) What happens with the lines parallel to the x-axis or parallel to the
y-axis?
(3) ϕ−1 (0) = y-axis. Is this fiber reduced?
∼
(4) ϕ induces an isomorphism C2 \ y-axis → C2 \ y-axis
Dominant morphisms. Let ϕ : X → Y be a morphism of affine vari-

eties, x a point of X and y := ϕ(x) its image in Y . Then ϕ∗ (my ) ⊆ mx , and
so ϕ∗ induces a local homomorphism
ϕ∗x : OY,y → OX,x .
(A homomorphism between local rings is called local if it maps the maximal
ideal into the maximal ideal.)
The next proposition tells us that, in a neighborhood of a point x ∈ X,
a morphism ϕ is uniquely determined by the local homomorphism ϕ∗x .
Proposition 3.33. (1) If ϕ, ψ : X → Y are two morphisms such

that ϕ(x) = ψ(x) and ϕ∗x = ψx∗ for some x ∈ X, then ϕ and ψ
coincide on every irreducible component of X passing through x.
∼
(2) If x ∈ X, y ∈ Y and if ρ : OY,y → OX,x is an isomorphism, then
there exist special open sets X 0 ⊆ X and Y 0 ⊆ Y containing x and
∼
y, respectively, and an isomorphism ψ : X 0 → Y 0 such that ψx∗ = ρ.
Proof. (1) Let R be a finitely generated reduced C-algebra and m ⊆ R
a maximal ideal. The canonical map µ : R → Rm is injective if and only if
m contains all minimal prime ideals of R. (In fact, ker µ = {r ∈ R | sr =
0 for some s ∈ R \ m}.)
Denote by X̄ ⊆ X the union of irreducible components passing through
x and by Ȳ ⊆ Y the union of irreducible components passing through ϕ(x).
Then ϕ(X̄) ⊆ Ȳ , because the image of an irreducible component of X is con-
tained in an irreducible component of Y (see Exercise 38). Thus we obtain a
morphism ϕ̄ : X̄ → Ȳ with the following commutative diagram of C-algebras
and homomorphisms which shows that ϕ̄ is completely determined by ϕ∗x :
⊆
O(Y ) −−−→ O(Ȳ ) −−−→ OȲ ,ϕ(x) = OY,ϕ(x)
  
ϕ∗ ϕ̄∗ ϕ∗
y y y x
⊆
O(X) −−−→ O(X̄ ) −−−→ OX̄,x = OX,x
(2) We can assume that all irreducible components of X pass through x and
all irreducible components of Y pass through y. Then
∼
O(Y ) ⊆ OY,y → OX,x ⊇ O(X).
Let h1 , . . . , hm ∈ O(Y ) be a set of generators and put fj := ρ(hj ). Then
we can find an element q ∈ O(X) \ mx such that fj ∈ O(X)q for all j,
i.e., ρ(O(Y )) ⊆ O(X)q . Now q = ρ( rs ) where r, s ∈ O(Y ), s ∈ / my and so
h := ρ(s)q = ρ(r) ∈ / mx . But this implies that ρ(O(Y )s ) = O(X)h . Hence,
∼
there is an isomorphism ψ : Xh → Ys with the required properties.
Definition 3.34. Let X, Y be irreducible affine varieties. A morphism
ϕ : X → Y is called dominant if the image is dense in Y , i.e. ϕ(X) = Y .
This is equivalent to the condition that ϕ∗ : O(Y ) → O(X) is injective (see
Proposition 3.31(2)).
It follows that every dominant morphism ϕ : X → Y induces a finitely
generated field extension ϕ∗ : C(Y ) ,→ C(X). If this is a finite field extension
of degree d := [C(X) : C(Y )] we will say that ϕ is a morphism of finite degree
2. MORPHISMS 67
∼
d. If d = 1, i.e. if ϕ∗ induces an isomorphism C(Y ) → C(X) then ϕ is called
a birational morphism.
Exercise 45. Let ϕ : C → C be a non-constant morphism. Then ϕ has finite
degree d, and there is a non-empty open set U ⊆ C such that #ϕ−1 (x) = d for all
x ∈ U.
There is a similar result as the second part of Proposition 3.33 saying
that affine varieties with isomorphic function fields are locally isomorphic.
The proof is similar as the proof above and will be left to the reader.
Proposition 3.35. Let X, Y be irreducible affine varieties and let
∼
ρ : C(Y ) → C(X) be an isomorphism. Then there exist special open sets
∼
X 0 ⊆ X and Y 0 ⊆ Y and an isomorphism ψ : X 0 → Y 0 such that ρ = ψ ∗ .
Products. If f is a function on X and h a function on Y then we denote
by f ·h the C-valued function on the product X ×Y defined by (f ·h)(x, y) :=
f (x) · h(y).
Proposition 3.36. The product X × Y of two affine varieties together
with the algebra
O(X × Y ) := C[f · h | f ∈ O(X), h ∈ O(Y )]
of C-valued functions is an affine variety. Moreover, the canonical homomor-
phism O(X) ⊗ O(Y ) → O(X × Y ), f ⊗ h 7→ f · h, is an isomorphism.
Proof. Let X ⊆ Cn and Y ⊆ Cm be closed subvarieties. Then X × Y ⊆
n+m
C is closed, namely equal to the zero set V(I(X) ∪ I(Y )). So it remains
to show that O(X × Y ) = C[x1, . . . , xn , y1, . . . , ym ]/I(X × Y ) is generated by
the products f · h and that f · h ∈ O(X × Y ) for f ∈ O(X) and h ∈ O(Y ).
But this is clear since x̄i = xi |X×Y = xi |X · 1 and ȳj = yj |X×Y = 1 · yj |Y , and
f |X · h|Y = (fh)|X×Y for f ∈ C[x1, . . . , xn ] and h ∈ C[y1, . . . , ym ].
For the last claim, we only have to show that the map O(X) ⊗ O(Y ) →
O(X × Y ), f ⊗ h 7→ f · h, is injective. For this, let (fi | i ∈ I) be a basis
of O(YP). Then every element s ∈ O(X) ⊗ O(Y ) can be uniquely P written as
s = finite si ⊗ fi . If s is in the kernel of the map, then P si (x)fi (y) = 0
for all (x, y) ∈ X × Y and so, for every fixed x ∈ X, si (x)fi is the zero
function on Y . This implies that si (x) = 0 for all x ∈ X and so si = 0 for all
i. Thus s = 0 proving the claim.
Example 3.20.
(1) The two projections prX : X × Y → X, (x, y) 7→ x, and prY : X ×
Y → Y , (x, y) 7→ y, are morphisms with comorphisms pr∗X (f ) = f · 1
and pr∗Y (h) = 1 · h.
(2) If ϕ : X → X 0 and ψ : Y → Y 0 are morphisms, then so is

ϕ × ψ : X × Y → X 0 × Y 0, (x, y) 7→ (ϕ(x), ψ(y)).
(3) Diagonal: ∆ : X → X × X, x 7→ (x, x) is a morphism, ∆(X) ⊆
X × X is a closed subset defined by the set {f · 1 − 1 · f | f ∈ O(X)},
∼
and X → ∆(X) is an isomorphism whose inverse is induced by the
projection prX .
(4) Graph: Let ϕ : X → Y be a morphism. Then
Γ(ϕ) := {(x, ϕ(x)) | x ∈ X} ⊆ X × Y
is a closed subset. Moreover, the projection prX induces an isomor-
∼
phism p : Γ(ϕ) → X and ϕ = prY ◦p−1 .
(5) Matrix multiplication: The composition of linear maps
µ : Hom(U, V ) × Hom(V, W ) → Hom(U, W ), (A, B) 7→ B ◦ A
P
is a morphism. Choosing coordinates we find µ∗ (zij ) = k yik xkj .
Exercise 46. Show that the ideal of the diagonal ∆(X) ⊆ X × X is generated
by the function f · 1 − 1 · f , f ∈ O(X) (see Example 3.20(3)).
Lemma 3.37. The projection prX : X × Y → X is an open morphism,

i.e. the image of an open set under prX is open.
Proof. It suffices to showPthat the image of a special open set U :=
(X × Y )g isSopen. Writing g = fi · hi with linearly independent hi one gets
prX (U ) = i Xfi and the claim follows.
Proposition 3.38. If X, Y are irreducible affine varieties then X × Y
is irreducible.
Proof. Assume that X × Y = A ∪ B with closed subsets A, B. Define
XA := {x ∈ X | {x} × Y ⊆ A} and XB := {x ∈ X | {x} × Y ⊆ B}.
Since Y is irreducible we see that X = XA ∪ XB . Now we claim that XA
and XB are both closed in X and so one of them equals X, say XA = X.
But then A = X × Y and we are done. To prove the claim we remark that
X \ XA = prX (X × Y \ A) which is open by Lemma 3.37 above.
S S
Corollary 3.39. If X = i Xi and Y = j Yj are the irreducible de-
S
compositions of X and Y , then X × Y = i,j Xi × Yj is the irreducible
decomposition of the product.
3. DIMENSION 69
3. Dimension
Definitions. If k is a field and A a k-algebra then a set a1, a2, . . . , an ∈ A
of elements from A are called algebraically independent over k if they do
not satisfy a non-trivial polynomial equation F (a1, a2, . . . , an ) = 0 where
F ∈ k[x1, . . . , xn ]. Equivalently, the canonical homomorphism of k-algebras
k[x1, . . . , xn ] → A defined by xi 7→ ai is injective.
In order to define the dimension of a variety we will need the concept of
transcendence degree tdegk K of a field extension K/k. It is defined to be the
maximal number of algebraically independent elements in K. We refer to [6,
Chap. 13, Sect. 8] for the basic properties of transcendental extensions.
Definition 3.40. Let X be an irreducible affine variety and C(X) its
field of rational functions. Then the dimension of X is defined by
dim X := tdegC C(X).
S
If X is reducible and X = Xi the irreducible decomposition (see 1) then
dim X := max dim Xi .
i
S
Finally, we define the local dimension of X in a point x ∈ X = Xi to be
dimx X := max dim Xi .
Xi 3x
Exercise 47. Let X be an affine variety.

(1) dim X is the maximal number of algebraically independent elements in
O(X).
(2) Assume that O(X) is generated by r elements. Then dim X ≤ r, and if
dim X = r then X ' Cr .
Exercise 48. The function x 7→ dimx X is upper semi-continuous on X. (This
means that for all α ∈ R the set {x ∈ X | dimx X < α} is open in X.)
Examples 3.21.
(1) We have dim Cn = n. More generally, if V is a complex vector space
of dimension n then dim X = n.
(2) If U ⊆ X is a special open subset which is dense in X, then dim U =
dim X.
(3) For affine varieties X, Y we have dim X × Y = dim X + dim Y .
Lemma 3.41. Let f ∈ C[x1, . . . , xn ] be a non-constant polynomial and
X := V(f ) ⊆ Cn its zero set. Then dim X = n − 1.
Proof. We can assume that f is irreducible and that the variable xn

occurs in f . Denote by x̄i ∈ O(X) = C[x1, . . . , xn ]/(f ) the restrictions of the
coordinate functions xi . Then C(X) = C(x̄1, x̄2 , . . . , x̄n ). Since
f (x̄1 , x̄2, . . . , x̄n ) = 0
we see that x̄n ∈ C(X) is algebraic over the subfield C(x̄1 , x̄2, . . . , x̄n−1 ).
Therefore, tdeg C(X) = tdeg C(x̄1 , x̄2, . . . , x̄n−1 ) ≤ n − 1. On the other hand,
the composition
res
C[x1, . . . , xn−1 ] ,→ C[x1, . . . , xn ] → O(X)
is injective, since the kernel is the intersection (f ) ∩ C[x1, . . . , xn−1 ] which is
zero. Thus, tdeg C(X) ≥ n − 1, and the claim follows.
The first part of the proof above, namely that dim V(f ) < n = dim Cn
has the following generalization.
Lemma 3.42. If X is irreducible and Y ( X a proper closed subset then
we have dim Y < dim X.
Proof. We can assume that Y is irreducible. If h1 , . . . , hm ∈ O(Y ) are
algebraically independent where m = dim Y , and hi = h̃i |Y for h̃1 , . . . , h̃m ∈
O(X) then h̃1 , . . . , h̃m are algebraically independent, too, and so dim X ≥
dim Y . If dim Y = dim X then every f ∈ O(X) is algebraic over C(h̃1 , . . . , h̃m ).
Choose f ∈ O(X) in the kernel of the restriction map, i.e. f |Y = 0. Then f
satisfies an equation of the form
f k + p1 f k−1 + · · · + pk−1 f + pk = 0
where pj ∈ C(h̃1 , . . . , h̃m ) and k is minimal. Multiplying this equation with
a suitable q ∈ C[h̃1 , . . . , h̃m ] we can assume that pj ∈ C[h̃1, . . . , h̃m ]. But this
implies that pk |Y = 0. Thus pk = 0 and we end up with a contradiction.
Example 3.22. We have dim X = 0 if and only if X is finite, and this is
equivalent to dimC O(X) < ∞. (This is clear: If X is irreducible of dimension
0 then C(X) is algebraic over C and so C = O(X) = C(X), and the claim
follows.)
Exercise 49. Let U ⊆ X be a dense open set. Then dim X \ U < dim X.
Lemma 3.43. Let X be an irreducible affine variety of dimension n. Then

there is a special open set U ⊆ X which is isomorphic to a special open set
of a hypersurface V(h) ⊆ Cn+1 .
3. DIMENSION 71
Proof. The field of rational functions C(X) has the form

C(X) = C(x1 , . . . , xn )[f ]
where f satisfies a minimal equation: f m + p1 f m−1 + · · · + pm = 0, pj ∈
C(x1, . . . , xn ). Multiplying with a suitable polynomial C[x1, . . . , xn ] we can
assume that all pj belong to C[x1, . . . , xn ]. Then the polynomial h := y m +
p1 y n−1 + · · · + pm ∈ C[x1, . . . , xn , y] is irreducible and defines a hypersurface
H := V(h) ⊆ Cn+1 whose field of rational functions C(H) is isomorphic to
C(X) by construction. Now the claim follows from Proposition 3.35.
Finite morphisms. Finite morphisms will play an important role in the

following. In particular, they will help us to “compare” an arbitrary affine
variety X with an affine space Cn of the same dimension by using the famous
Normalization Lemma of Noether.
Definition 3.44. Let A ⊆ B be two rings. We say that B is finitePover A
if B is a finite A-module, i.e.there are b1, . . . , bs ∈ B such that B = j Abj .
A morphism ϕ : X → Y between two affine varieties is called finite if
O(X) is finite over ϕ∗ (O(Y )).
If A ⊆ B ⊆ C are rings such that B is finite over A and C is finite over B,
then C is finite over A. In particular, if ϕ : X → Y and ψ : Y → Z are finite
morphisms then the composition ψ ◦ ϕ : X → Z is finite, too. Another useful
remark is the following: If ϕ : X → Y is finite and X 0 ⊆ X, Y 0 ⊆ Y closed
subsets such that ϕ(X 0 ) ⊆ Y 0 then the induced morphism ϕ0 : X 0 → Y 0 is
also finite.
Example 3.23. Typical examples of finite morphisms are the ones given
in Example 3.17, namely ϕ : C → C = V(y 2 − x3 ) ⊆ C2 and ψ : C →
D = V(y 2 − x2 − x3) ⊆ C2. In both cases, the morphisms are the so-called
normalizations, a concept which we will discuss later.
On the other hand, the inclusion of a special open set Xf ,→ X is not
finite if f is neither invertible nor zero.
Exercise 50. Every non-constant morphism ϕ : C → C is finite, and the same
holds for the non-constant morphisms ψ : C∗ → C∗ .
The basic geometric property of a finite morphism is given in the next
proposition.
Proposition 3.45. Let ϕ : X → Y be a finite morphism. Then ϕ is
closed and has finite fibers.
Proof. If y ∈ Y then ϕ−1 (y) = VX (ϕ∗ (my )) (see 2). If ϕ−1 (y) 6= ∅ then
the induced morphism ϕ−1 (y) → {y} is finite, too, and so O(ϕ−1 (y)) is a
finite dimensional C-algebra. Thus, the fiber ϕ−1 (y) is finite (Example 3.22)
proving the second claim.
For the first claim it suffices to show that ϕ(X) = ϕ(X). Hence we
can assume that ϕ(X) = Y , i.e. that ϕ∗ : O(Y ) → O(X) is injective. If
ϕ−1 (y) = ∅ then O(X)my = O(X). (We identify my with its image ϕ∗ (my ).)
The Lemma of Nakayama (see the following Lemma 3.46) now implies that
(1+a)O(X) = 0 for some a ∈ my which is a contradiction since 1+a 6= 0.
Lemma 3.46 (Lemma of Nakayama). Let R be a ring, a ⊆ R an ideal
and M a finitely generated R-module. If aM = M then there is an element
a ∈ a such that (1 + a)M = 0. In particular, if M is torsionfree and a 6= R
then M = 0.
P P
Proof. Let M = kj=1 Rmj . Then mi = j aij mj for all i where aij ∈ a.
If A denotes the k × k-matrix (aij )i,j and m the column vector (m1, . . . , mk )t
this means that m = A · m. Thus (E − A)m = 0, and so det(E − A)mj = 0
for all j. But
 
1 − a11 −a12 · · ·
det(E − A) = det  −a21 1 − a22 · · ·  = 1 + a where a ∈ a.
.. ..
. .
Exercise 51. Let X be an affine variety and x ∈ X. Assume that f1 , . . . , fr ∈
mx generate the ideal mx modulo m2x, i.e., mx = (f1, . . . , fr ) + m2x . Then {x} is
an irreducible component of VX (f1 , . . . , fr ). (Hint: If C ⊆ VX (f1, . . . , fr ) is an
irreducible component containing x and m ⊆ O(C) the maximal ideal of x then
m2 = m. Hence m = 0 by the Lemma of Nakayama above.)
Exercise 52. Let ϕ : X → Y be a finite surjective morphism. Then dim X =
dim Y .
S
Exercise 53. Let X be an affine variety and X = i Xi the irreducible de-
composition. A morphism ϕ : X → Y is finite if and only if ϕ|Xi : Xi → Y is finite
for all i.
The following easy lemma will be very useful later on .
Lemma 3.47. Let A ⊆ B be rings and b ∈ B. Assume that b satisfies an
equation of the form
bm + a1bm−1 + a2bm−2 + · · · + am = 0
where a1, a2, . . . , am ∈ A. Then the subring A[b] ⊆ B is finite over A.
3. DIMENSION 73
Proof. It follows from the equation satisfied by b that for N ≥ m we

have
(3.3) bN = −a1bN −1 − a2bN −2 − · · · − am bN −m ,
P
and so, by induction, that A[b] = m−1 i
i=0 Ab .
The next result is usually called the “Normalization Lemma”. It is due
to Emmy Noether, but was first formulated, in a special case, by David
Hilbert.
Theorem 3.48 (Noether’s Normalization Lemma). Let K be an infi-
nite field and A a finitely generated K-algebra. Then there are algebraically
independent elements a1, . . . , an ∈ A such that A is finite over K[a1, . . . , an ]
Proof. We proceed by induction on the number m of generators of A
as a K-algebra. If m = 0 then A = K and there is nothing to prove. If
A = K[b1, . . . , bm ] and if b1 , . . . , bm are algebraically independent, we are
done, too. So let’s assume that F (b1, . . . , bm ) = 0 where F ∈ K[x1, . . . , xm]
is a non-zero polynomial. We can also assume that xm occurs in F . Write
X
F = αr1 ,r2 ,...,rm xr11 xr22 · · · xrmm
r1 ,r2 ,...,rm
and put r := max{r1 + r2 + · · · + rm | αr1 ,r2 ,...,rm 6= 0}. Substituting xj =

x0j + γj xm for j = 1, . . . , m − 1 and we find
X rm−1
(3.4) F = ( αr1 ,...,rm γ1r1 · · · γm−1 ) xrm + H(x01 , . . . , x0m−1 , xm )
r1 +r2 +···+rm =r
where xm occurs in H with anP exponent < r. Since K is infinite we can find
r1 rm−1
γ1 , . . . , γm−1 ∈ K such that r1 +···+rm =r αr1 ,...,rm γ1 · · · γm−1 6= 0. Setting
b0j := bj − γj bm for j = 1, . . . , m − 1 we have A = K[b01, b02 , . . . , b0m−1 , bm ].
Now equation (3.4) implies that bm satisfies an equation of the form (3.3),
hence A is finite over K[b01, . . . , b0m−1 ] by Lemma 3.47, and the claim follows
by induction.
Remark 3.49. The proof above shows the following. If A = K[b P 1, . . . , bm]
then there are a number n ≤ m and n linear combinations ai := j γij bj ∈ A
such that a1, . . . , an are algebraically independent over K and that A is finite
over K[a1, . . . , am].
A first consequence is the following result.
Proposition 3.50. Let X be an affine variety of dimension n. Then
there is a finite surjective morphism ϕ : X → Cn .
Proof. It follows from the Normalization Lemma (Theorem 3.48) that

there exist f1 , . . . , fn ∈ O(X) such that O(X) is finite over the subring
C[f1, . . . , fn ]. It follows that dim X = n (see Exercise 47) and that the mor-
phism ϕ = (f1 , . . . , fn ) : X → Cn is finite and surjective (Proposition 3.45).

This result can be improved, using Remark 3.49 above.
Proposition 3.51. Let X ⊆ Cm be a closed subvariety of dimension
n ≤ m. Then there is a linear projection λ : Cm → Cn such that λ|X : X → Cn
is finite and surjective.
In fact, more is true: There is an open dense set U ⊆ Hom(Cm , Cn ) such
that the proposition above holds for any λ ∈ U . We will not give a proof here
since it does not follow immediately from our previous results.
Exercise 54. Let f1 , f2, . . . , fm ∈ C[x1, . . . , xn ] be non-constant homogeneous
polynomials and put A := C[f1 , f2, . . . , fm ]. Then the following statements are
equivalent:
(i) C[x1, . . . , xn ]/(f1, f2, . . . , fm ) is a finite dimensional algebra;
(ii) There is a k ∈ N such that (x1, x2, . . . , xn )k ⊆ (f1 , f2, . . . , fm );
(ii) C[x1, . . . , xn ] is finite over A.
(Remark: The fact, that the fi ’s are homogeneous is essential!)
Exercise 55. Assume that the morphism ϕ : Cn → Cm is given by homoge-
neous polynomials f1 , · · · , fm . If ϕ−1 (0) is finite then ϕ−1 (0) = {0} and ϕ is a
finite morphism.
(Hint: This follows immediately from the previous exercise.)
Exercise 56. Let X ⊆ Cn be a cone and λ : Cn → Cm a linear map. If
X ∩ ker λ = {0} then λ|X : X → Cm is finite. Moreover, the set of linear maps
λ : Cn → Cm such that λ|X is finite is open in Hom(Cn , Cm).
Krull’s principal ideal theorem. We have seen in Lemma 3.41 that

the dimension of a hypersurface V(f ) ⊆ Cn is equal to n − 1, i.e.
codimCn V(f ) = 1,
where the codimension of a closed subvariety Y ⊆ X is defined by
codimX Y := dim X − dim Y.
We want to generalize this to arbitrary affine varieties X. First we prove a
converse of Lemma 3.47.
3. DIMENSION 75
Lemma 3.52. Let A ⊆ B be rings. Assume that A is Noetherian and that

B is finite over A. Then every b ∈ B satisfies an equation of the form
bm + a1bm−1 + a2bm−2 + · · · + am = 0
where a1, a2, . . . , am ∈ A.
Proof. Since A is Noetherian the subalgebra A[b] ⊆ B is finite over A.
Therefore, the sequence A ⊆ A + Ab ⊆ A + Ab + Ab2 ⊆ · · · ⊆ A + Ab +
· · · + Abk ⊆ · · · becomes stationary. Hence, there is a m ≥ 1 such that
bm ∈ A + Ab + · · · + Abm−1 .
Exercise 57. Let r ∈ C(x1, . . . , xn) satisfy an equation of the form
rm + p1rm−1 + · · · + pm = 0 where pj ∈ C[x1, . . . , xn ].
Then r ∈ C[x1, . . . , xn]. In particular, if A ⊆ C(a1, . . . , an ) is a subalgebra which
is finite over C[a1, . . . , an ] then A = C[a1, . . . , an ].
Lemma 3.53. Let A be a C-algebra without zero divisors and K its field
of fractions. Let a1, . . . , an ∈ A be algebraically independent elements such
that A is finite over C[a1, . . . , an ] and denote by N : K → C(a1, . . . , an ) the
norm. Then
(1) N (A) ⊆ C[a1, . . . , an ]; p p
(2) For all a ∈ A we have Aa ∩ C[a1, . . . , an ] = C[a1, . . . , an ]N (a).
Proof. Let L/K be a finite field extension containing all the conjugates
a(1) := a, a(2), . . . , a(r) of a where r = [K : C(a1 , . . . , an )]. Since a belongs to
an algebra which is finite over C[a1, . . . , an ], namely A, the same holds for all
a(j). Thus, every a(j) satisfies an equation with coefficients in C[a1, . . . , an ]
and leading coefficient 1 (Lemma 3.52). This implies by Lemma 3.47 that the
subalgebra Ã := C[a1, . . . , an ][a(j)] ⊆ L is finite over C[a1, . . . , an ] and con-
tains all a(j). Therefore, N (a) = a(1)a(2) · · · a(r) belongs to Ã ∩ C(a1 , . . . , an )
which is equal to C[a1, . . . , an ] by the exercise above. This proves the first
claim.
Now we have
Y
(t − a(j) ) = tr + h1 tr−1 + · · · + hr−1 t + hr
j
where hj ∈ Ã ∩ C(a1, . . . , an ) = C[a1, . . . , an ] and hr = (−1)r N (a). It follows

that N (a) = ab where b = (−1)r−1 (ar−1 + h1 ar−2 + · · · + hr−1 ) ∈ A and so
N (a) ∈ Aa. Thus, C[a1, . . . , an ]N (a) ⊆ Aa ∩ C[a1, . . . , an].
p
In order to see that Aa ∩ C[a1, . . . , an ] ⊆ C[a1, . . . , an ]N (a) we choose
an element sa ∈ Aa ∩ C[a1, . . . , an ]. Then N (sa) =p (sa)r , and since N (sa) =
N (s)N (a) ∈ C[a1, . . . , an ]N (a) we finally get sa ∈ C[a1, . . . , an ]N (a).
Theorem 3.54 (Krull’s Principal Ideal Theorem). Let X be an irre-
ducible affine variety and f ∈ O(X), f 6= 0. Assume that VX (f ) is non-
empty. Then every irreducible component of VX (f ) has codimension 1 in X.
In particular, dim VX (f ) = dim X − 1.
Proof. Let VX (f ) = C1 ∪ C2 ∪ · · · ∪ Cr be the irreducible decompo-
sition. Choose an h ∈ O(X) vanishing on C2 ∪ C3 ∪ · · · ∪ Cr which does
not vanish on C1 . Then VXh (f ) = C1 ∩ Xh is irreducible. Thus, it suf-
fices to consider the case where VX (f ) ⊆ X is irreducible. By the Nor-
malization Lemma (Theorem 3.48) there is a finite surjective morphism
ϕ : X → Cn , n = dim X. By Lemma 3.53(2) we get ϕ(VX (f )) = V(N (f )),
and so dim VX (f ) = dim V(N (f )) = n − 1 (see Lemma 3.41).
It is easy to see that this result also holds for equidimensional varieties
(i.e. varieties X where all irreducible components have the same dimension)
if f is a non-zero divisor. For a general X and a non-zero divisor f ∈ O(X),
we can only say that every irreducible component of VX (f ) has dimension
≤ dim X − 1.
A first consequence is the following result.
Proposition 3.55. Let X be an irreducible variety and f1 , f2 , . . . , fr ∈
O(X). If VX (f1, . . . , fr ) is non-empty then every irreducible component C of
VX (f1 , . . . , fr ) has dimension dim C ≥ dim X − r.
Proof. We proceed by induction on dim X. Define Y := VX (f1 ), and let
Y = Y1 ∪ · · · ∪ Ys be the decomposition into irreducible components. Then
[
VX (f1 , · · · , fr ) = VYj (f2 , . . . , fr )
j
Since dim Yj = dim X −1 for all j we see, by induction, that every irreducible
component of VYj (f2 , . . . , fr ) has dimension ≥ (dim X −1)−(r−1) = dim X −
r, and the claim follows.
Exercise 58. Let X be an affine variety and f ∈ O(X) a non-zero divisor.
For any x ∈ VX (f ) we have dimx VX (f ) = dimx X − 1. (Hint: If f is a non-zero
divisor, then f is non-zero on every irreducible component Xi of X and so VXi (f )
is either empty or every irreducible component has codimension 1. Now the claim
follows easily.)
3. DIMENSION 77
Another consequence of Krull’s Principal Ideal Theorem is the following

which gives an alternative definition of the dimension of a variety.
Proposition 3.56. Let X be an irreducible variety and Y $ X a closed
irreducible subset. Then there is a strictly decreasing chain of length n :=
dim X,
Xn = X % Xn−1 % · · · % Xd = Y % · · · % X1 % X0
of irreducible closed subsets Xj . In particular, dim X equals the length of a
maximal chain of irreducible closed subsets.
Proof. By induction, we only have to show that Y is contained in an
irreducible hypersurface H ⊆ X. Let f ∈ I(Y ) be a non-zero function. Then
X ⊇ VX (f ) ⊇ Y and so Y is contained in an irreducible component of VX (f )
which all have codimension 1 by Theorem 3.54.
Remark 3.57. This result allows to define the dimension dim A of a C-
algebra A as the maximal length of a chain of prime ideal p0 ⊆ p1 ⊆ · · · ⊆
pm ⊆ A. If A is finitely generated then dim A is finite, and every maximal
p
chain has length dim A. Moreover, dim A = dim Ared where Ared := A/ (0),
and so dim A = dim X where X is an affine variety with coordinate ring
isomorphic to Ared.
We also see that for a variety X and a point x ∈ X we have dimx X =
dim OX,x.
Corollary 3.58. Let A be a finitely generated C-algebra and let a ∈ A
be a non-zero divisor. Then dim A/Aa ≤ dim A − 1, and equality holds if Ared
is a domain.
Proof. Put Ā := A/(a) and denote by a0 ∈ A pred the image of a. Then
0 0
a is a non-zero divisor in Ared and
p so dim Ared/ (a ) ≤ dim Ared − 1 by
Theorem 3.54. Since Āred ' Ared/ (a0) we finally get dim Ā = dim Āred ≤
dim Ared − 1 = dim A − 1
Decomposition Theorem and dimension formula. Let ϕ : X → Y

be a dominant morphism where X, Y are both irreducible. We want to show
that the dimension of a non-empty fiber ϕ−1 (y) is always ≥ dim X − dim Y
and that we have equality on a dense open set of Y . A crucial step is the
following Decomposition Theorem for a morphism.
Theorem 3.59. Let X and Y be irreducible varieties and ϕ : X → Y a
dominant morphism. There is a non-empty special open set U ⊆ Y and a
factorization of ϕ of the form

ρ
ϕ−1 (U ) / U × Cr
LLL
LLL prU
ϕ LLLL
L%
U
where ρ is a finite surjective morphism and r := dim X −dim Y . In particular,
the fibers ϕ−1 (y) = ρ−1 ({y} × Cr ) have the same dimension for all y ∈ U ,
namely dim X − dim Y .
Remark 3.60. We will see later in Proposition 3.62 that the fibers ϕ−1 (y)
for y ∈ U are equidimensional, i.e., all irreducible components have the same
dimension, namely dim X − dim Y .
Proof. Since ϕ is dominant we will regard O(Y ) as a subalgebra of
O(X). Let K = C(Y ) be the quotient field of O(Y ) and put A := K · O(X) ⊆
C(X), the K-algebra generated by K and O(X). Then A is finitely generated
over K and so we can find algebraically independent elements h1 , . . . , hr ∈ A
such that A is finite over K[h1, . . . , hr ] (Theorem 3.48). It follows that r =
dim X − dim Y .
We claim that there is an f ∈ O(Y ) such that hi = afi with ai ∈ O(X)
for all i and that O(Xf ) = O(X)f is finite over O(Yf )[h1 , . . . , hr ]. The first
statement is clear, and we can therefore assume that h1, . . . , hr ∈ O(X).
For the second statement, let b1, . . . , bs be generators of A over K[h1, . . . , hr ].
Multiplying with a suitable element of O(Y ) ⊆ K we can first assume that
bj ∈ O(X) and then, by adding more elements if necessary, that b1 , . . . , bs gen-
P (ij) (ij)
erate O(X) as a C-algebra. Now bi bj = k ck bk where ck ∈ K[h1 , . . . , hr ].
(ij)
Thus we can find an f ∈ O(Y ) such that f · ck ∈ O(Y )[h1, . . . , hr ]. It follows
that X
O(Yf )[h1, . . . , hr ] bj ⊆ O(X)f = O(Xf )
j
is a subalgebra containing O(X), hence is equal to O(Xf ), and the claim
follows.
Setting U := Yf we get ϕ−1 (U ) = Xf and obtain a morphism
ρ = ϕ × (h1, . . . , hr ) : Xf → Yf × Cr , x 7→ (ϕ(x), h1(x), . . . , hr (x))
which satisfies the requirements of the proposition.
The last statement is clear (see Exercise 52).
Exercise 59. Work out the decomposition of Theorem 3.59 in the case of the
following morphisms:
3. DIMENSION 79
(1) ϕ : M2 → M2 , ϕ(A) 2
:= A .
a b
(2) ϕ : SL2 → C3 , ϕ( ) := (ab, ad, cd) (see Exercise 43).
c d
What is the degree of the finite morphism ρ in each case?
Corollary 3.61. If ϕ : X → Y is a morphism, then there is a set U ⊆

ϕ(X) which is open and dense in ϕ(X).
Proof. If X is irreducible, this is an immediate consequence of Theo-
rem 3.59 above. In general, we apply this proposition to every irreducible
component of X, and use Exercise 23.
Proposition 3.62. Let X and Y be irreducible varieties and ϕ : X → Y
a dominant morphism. If y ∈ ϕ(X) and C is an irreducible component of the
fiber ϕ−1 (y) then
dim C ≥ dim X − dim Y.
Proof. Set m := dim Y and let ψ : Y → Cm be a finite surjective
morphism (Theorem 3.48). If we denote by ϕ̃ : X → Cm the composition
ψ ◦ ϕ, then every fiber of ϕ̃ is a finite union of fibers of ϕ. Hence it suf-
fices to prove the claim for the morphism ϕ̃ = (f1 , . . . , fm ) : X → Cm . If
a = (a1 , . . . , am ) ∈ ϕ̃(X) then ϕ̃−1 (a) = VX (f1 − a1 , f2 − a2, . . . , fm − am),
and the claim follows from Proposition 3.55, a consequence of Krull’s Prin-
cipal Ideal Theorem.
One might believe that the two propositions above imply that for any
morphism ϕ : X → Y the function y 7→ dim ϕ−1 (y) is upper-semicontinuous
(see below). This is not true as one can show by examples. However, a famous
theorem of Chevalley says that the function x 7→ dimx ϕ−1 (ϕ(x)) is upper-
semicontinuous on X. The proof is quite involved and we will not present it
here.
Example 3.24. Consider the morphism ϕ : C2 → C2 given by (x, y) 7→
(x, xy). It is easy to see that the image ϕ(C2 ) is not locally closed in C2 and
that the map a 7→ dim ϕ−1 (a) is not upper-semicontinuous.
Constructible sets. Recall that a subset A ⊆ X of a variety X is
called locally closed if A is the intersection of an open and a closed subset,
or, equivalently, if A is open in its closure Ā. We have seen in examples
that images of morphisms need not to be locally closed in general. However,
we will show that images of morphisms are always “constructible” in the
following sense.
Definition 3.63. A subset C of an affine variety X is called constructible

if it is a finite union of locally closed subsets.
Exercise 60.
(1) Finite unions, finite intersections and complements of constructible sets
are again constructible.
(2) If C is constructible, then C contains a set U which is open and dense in
C̄.
Proposition 3.64. If ϕ : X → Y is a morphism then the image of a
constructible subset is again constructible.
Proof. Since every open set is the union of finitely many special open
sets it suffices to show, in view of the exercise above, that the image of a
morphism is constructible. By Corollary 3.61 there is a dense open set U ⊆
ϕ(X) contained in the image ϕ(X). Then the complement Y 0 := ϕ(X) \ U
is closed and dim Y 0 < dim Y (Exercise 49). By induction on dim ϕ(X), we
can assume that the claim holds for the morphism ϕ0 : X 0 := ϕ−1 (Y 0) → Y 0
induced by ϕ. But then ϕ(X) = U ∪ ϕ0(X 0 ) and we are done.
Degree of a morphism. Recall that a dominant morphism ϕ : X → Y
between irreducible varieties is called of finite degree d if dim X = dim Y
and d = [C(X) : C(Y )] (see the section on dominant morphisms). This has
the following geometric interpretation.
Proposition 3.65. Let X, Y be irreducible affine varieties and ϕ : X →
Y a dominant morphism of finite degree d. Then there is a dense open set
U ⊆ Y such that #ϕ−1 (y) = d for all y ∈ U .
Proof. We have C(X) = C(Y )[r] where r satisfies the equation
rd + a1rd−1 + · · · + ad = 0.
Replacing Y and X by suitable special open sets Yf and Xf (f ∈ O(Y )) we
can assume that
(1) O(X) is finite over O(Y ) (Theorem 3.59);
(2) r ∈ O(X);
(3) ai ∈ O(Y );
(4) O(X) = O(Y )[r].
This implies that
M
d−1
∼
O(X) = O(Y )rj ← O(Y )[t]/(td + a1 td−1 + · · · + ad ).
j=0
4. TANGENT SPACES AND DIFFERENTIALS 81
and so, for every y ∈ Y , we get

O(X)/O(X)my ' C[t]/(td + a1(y)td−1 + · · · + ad (y)).
This means that the number of elements in the fiber ϕ−1 (y) is equal to the
number of different solutions of the equation
(3.5) td + a1(y)td−1 + · · · + ad (y) = 0.
Now let D be the discriminant of an equation of degree d (see Example 3.2)
and define f (y) := D(a1 (y), . . . , ad(y)). Then f ∈ O(Y ), and f (y) 6= 0 if
and only if equation (3.5) has d different solutions, or, equivalently, the fiber
ϕ−1 (y) has d points. Thus, the special open set U := Yf ⊆ Y has the required
property.
Remark 3.66. One can show that the open set U constructed in the
proof has the property that the morphism ϕ−1 (U ) → U is an unramified
covering with respect to the C-topology.
Exercise 61. What is the degree of the morphism Mn → Mn given by
A 7→ Ak ?
Exercise 62. Let ϕ : X → Y be a dominant morphism where X and Y are
both irreducible. If there is an open dense set U ⊆ X such that ϕ|U is injective,
then ϕ is birational.
4. Tangent Spaces and Differentials

Zariski tangent space. A tangent vector δ in a point x0 of an affine
variety X is “rule” to differentiate regular functions, i.e., it is a C-linear map
δ : O(X) → C satisfying
(4.6) δ(f · g) = f (x0) δ(g) + g(x0 ) δ(f ) for all f, g ∈ O(X).
Such a map is called a derivation of O(X) in x0. It follows that δ(f n ) =
nf n−1 (x0)δ(f ) and so, for any polynomial F = F (y1, . . . , ym ), we get
Xm
∂F
δ(F (f1, . . . , fm )) = (f1 (x0 ), . . . , fm (x0)) δ(fj ).
j=1
∂yj
This implies that a derivation in x0 is completely determined by its values

on a generating set of the algebra O(X). Moreover, a linear combination of
derivations in x0 is again a derivation in x0. As a consequence, the derivations
in x0 form a finite dimensional subspace of Hom(O(X), C).
Definition 3.67. The Zariski tangent space Tx0 X of a variety X in a

point x0 is defined to be the set of all tangent vectors in x0:
Tx0 X := Derx0 (O(X)) := {δ : O(X) → C | δ a C-linear derivation in x0 }.
Tx0 X is a finite dimensional linear subspace of Hom(O(X), C).
Exercise 63. Let δ ∈ TxX be a tangent vector in x.
(1) δ(c) = 0 for every constant c ∈ O(X).
δf
(2) If f ∈ O(X) is invertible, then δ(f −1 ) = − .
f (x)2
Example 3.25. If X = Cn and a = (a1, . . . , an ) ∈ Cn then

M ∂
TaCn = C
∂xi
i a

∂ ∂f
where ∂xi (f ) := ∂x i
(a). Thus we have a canonical isomorphism Ta Cn ' Cn
a
by identifying δ ∈ Dera(C[x1, . . . , xn ]) with (δx1, . . . , δxn) ∈ Cn .
More generally, if V is a finite dimensional vector space and x0 ∈ V we
define, for every v ∈ V , the tangent vector ∂v,x0 : O(V ) → C in x0 by

f (x0 + tv) − f (x0)
∂v,x0 (f ) := ,
t t=0
∼
and thus obtain a canonical isomorphism V → Tx0 V , for every x0 ∈ V .
Let δ ∈ TxX be a tangent vector. Since O(X) = C ⊕ mx we see that δ
is determined by its restriction to mx . Moreover, formula (4.6) shows that δ
vanishes on m2x . Hence, δ induces a linear map δ̄ : mx /m2x → C.
Lemma 3.68. Given an affine variety X and a point x ∈ X there is a
canonical isomorphism
∼
Tx X → Hom(mx /m2x , C)
given by δ 7→ δ̄ := δ|mx .
Proof. We have already seen that δ 7→ δ̄ is injective. On the other hand,
let C ⊆ mx be a complement of m2x so that O(X) = C ⊕ C ⊕ m2x . If λ : C → C
is linear then one easily sees that the extension of λ to a linear map δ on
O(X) by putting δ|C⊕m2x = 0 is a derivation in x.
Exercise 64. The canonical homomorphism O(X) → OX,x induces an iso-
∼
morphism mx/m2x → m/m2 where m ⊆ OX,x is the maximal ideal.
If U = Xf ⊆ X is a special open set and x ∈ U then TxU = TxX in a

canonical way. In fact, a derivation δ 0 of O(U ) induces a derivation δ of O(X)
by restriction: δ(h) := δ 0(h|U ), and every derivation δ of O(X) “extends”
to a derivation δ 0 of O(U ) = O(X)f by setting δ 0( fhm ) = (−m) f m+1 δh
(see
Exercise 63). The same result follows from Exercise 64 using Lemma 3.68.
Exercise 65. If Y ⊆ X is a closed subvariety and x ∈ Y then dim TxY ≤
dim Tx X.
(Hint: The surjection O(X) → O(Y ) induces a surjection mx,X /m2x,X → mx,Y /m2x,Y .)
Proposition 3.69. dim Tx X ≥ dimx X.

Proof. If C ⊆ X is an irreducible component passing through x we
have dim TxC ≤ dim TxX (Exercise 65). Thus we can assume that X is
irreducible. Choose f1 , . . . , fr ∈ mx such that the residue classes modulo m2x
form a basis of mx /m2x , hence r = dim TxX, by Lemma 3.68. Since the zero
set VX (f1 , . . . , fr ) has {x} as an irreducible component (see Exercise 51) it
follows from Proposition 3.55 that
0 = dim{x} ≥ dim X − r = dim X − dim Tx X.
Hence the claim.
∼
Proposition 3.70. There is a canonical isomorphism T(x,y)X × Y →
Tx X ⊕ Ty Y where x ∈ X and y ∈ Y .
Proof. Every derivation δ of O(X × Y ) in (x, y) induces, by restriction,
derivations δX of O(X) in x and δY of O(Y ) in y. This defines a linear map
T(x,y)X × Y → Tx X ⊕ Ty Y which is injective, because δ(f · h) = δX f · h(y) =
f (x) · δY h for f ∈ O(X) and h ∈ O(Y ).
In order to see that the map is surjective we first claim that given two
derivations δ1 ∈ Tx X and δ2 ∈ Ty Y there is a unique linear map δ : O(X ×
Y ) → C such that δ(f · h) = δ1f · h(y) = f (x) · δ2h. This follows from
Proposition 3.36 and the universal property of the tensor product. Now it is
easy to see that this map δ is a derivation in (x, y) and that δX = δ1 and
δY = δ2.
Tangent spaces of subvarieties. Let X ⊆ V be a closed subvariety

of the vector space V and x0 ∈ X. If δ ∈ Tx0 V = V is a tangent vector
which vanishes on I(X) = ker(res : O(V ) → O(X)) then the induced map
δ̄ : O(X) → C is a derivation in x0 , and vice versa. Thus we have the following
result.
Proposition 3.71. If X ⊆ V is a closed subvariety and x0 ∈ X then

Tx0 X = {δ ∈ Tx0 V | δ(f ) = 0 for all f ∈ I(X)} ⊆ Tx0 V = V.
More explicitly, let V = Cn and assume that the ideal I(X) is generated by
f1 , . . . , fs ∈ C[x1, . . . , xn ]. Then, for x0 ∈ X, we get
X
n
∂fi
n
Tx0 X = {a = (a1, . . . , an ) ∈ C | aj (x0 ) = 0 for i = 1, . . . , s}.
j=1
∂xj
In particular,

∂fi
dim Tx0 X = n − rk (x0 ) .
∂xj i,j

∂fi
The matrix (x0) is called the Jacobian matrix at the point x0 and
∂xj i,j
will be denoted by Jac(f1, . . . , fs )(x0).
Example 3.26. Consider the plane curve C = V(y 2 − x3) ⊆ C2 . Then
I(C) = (y 2 − x3) and so the tangent space in an arbitrary point x0 = (a, b) ∈
C is given by T(a,b)C = {(u, v) ∈ C2 | −3a2 u + 2bv = 0}. Since (a, b) = (t2, t3)
for some t ∈ C we get

2

C " # for t = 0,
T(t2 ,t3 )C = 2

C 3t for t 6= 0.
Exercise 66. Calculate the tangent spaces of the plane curves C1 := V(y −x2 )
and C2 = V(y 2 − x2 − x3) in arbitrary points (a, b).
Remark 3.72. Consider the C-algebra C[ε] := C[t]/(t2) called the alge-
bra of dual numbers. By definition, we have C[ε] = C ⊕ Cε and ε2 = 0. If
X is an affine variety and ρ : O(X) → C[ε] an algebra homomorphism, then
ρ is of the form ρ = evx ⊕δx ε for some x ∈ X where evx is the evaluation
map at x and δx a derivation in x, i.e., ρ(f ) = f (x) + δx(f )ε. Conversely, if
δx is a derivation in x then ρ := evx ⊕δxε is an algebra homomorphism. If
X = V is a vector space, then the homomorphisms ρ : O(V ) → C[ε] are in
one-to-one correspondence with the elements of V ⊕ V ε. In fact, there are
canonical bijections
∼ ∼
AlgC (O(V ), C[ε]) → Hom(V, C[ε]) → V ⊕ V ε,
and the inverse map associates to x+vε ∈ V ⊕V ε the algebra homomorphism

ρ : f 7→ f (x + vε). Since
f (x + vε) = f (x) + ∂v,x f ε
it follows again from the above that Tx V can be canonically identified with
V . This formula is very useful for calculating tangent spaces as we will see
below.
Example 3.27. (a) The tangent space of GLn at E is the space of all
n × n-matrices and the tangent space of SLn at E ∈ SLn is the subspace of
traceless matrices:
TE SLn = sln := {A ∈ Mn | tr A = 0} ⊆ TE GLn = gln := Mn .
In fact, I(SLn ) = (det −1), and an easy calculation shows that det(E +Aε) =
1+tr(A)ε which implies, by Proposition 3.71, that A ∈ Mn belongs to TE SLn
if and only if tr A = 0.
(b) Next we look at the orthogonal group On := {A ∈ Mn | AAt = E}. As
a closed subset On is defined by n+1 quadratic equations and so dim On ≥
2 n+1
n
2
n − 2 = 2 . On the other hand, we have
(E + Xε)(E + Xε)t = E + (X + X t )ε
which shows that TE On ⊆ {X ∈ Mn | X skewsymmetric}. Since this space
has dimension n2 and since dimE On = dim On (Exercise 22) it follows from
Proposition 3.69 that
TE On = TE SOn = son := {X ∈ Mn | X skewsymmetric}.
Exercise 67. If X, Y ⊆ Cn are closed subvarieties and z ∈ X ∩ Y then
Tz (X ∩ Y ) ⊆ Tz X ∩ Tz Y ⊆ Cn .
Nonsingular varieties. We have seen in Proposition 3.69 that for every

point x of an affine variety X one has dim TxX ≥ dimx X. We will show now
that equality holds in an open set and we will characterize these points.
Definition 3.73. The variety X is called nonsingular or smooth in
x ∈ X if dim Tx X = dimx X. Otherwise it is singular in x. The variety is
called nonsingular or smooth if it is nonsingular in every point. We denote
by Xsing the set of singular points of X.
Example 3.28. Let H := V(f ) ⊆ Cn be a hypersurface where f ∈
C[x1, . . . , xn ] is square-free and non-constant, and so I(H) = (f ). Then the
tangent space in a point x0 ∈ H is given by

X ∂f
Tx0 X := {a = (a1, . . . , an ) | ai (x0) = 0},
i
∂x i
and so
∂f ∂f ∂f
Hsing = V(f, , ,··· , ) ⊆ H.
∂x1 ∂x2 ∂xn
It follows that Hsing is a proper closed subset whose complement is dense.
∂f
(This is clear for irreducible hypersurfaces since a non-zero derivative ∂x i
∂f ∂f
cannot be a multiple of f and so V(f, ∂x 1
, · · · , ∂xn
) is a proper closed sub-
set of V(f ). This implies that every irreducible component of H contains
a non-empty open set of nonsingular points which does not meet the other
components, and the claim follows.)
It is also interesting to remark that a common point of two or more
irreducible components of H is always singular. We will see that this is true
in general (Corollary 3.80).
Proposition 3.74. Let X be an irreducible affine variety. Then the set
Xsing of singular points of X is a proper closed subset of X whose complement
is dense.
Proof. We can assume that X is an irreducible closed subvariety of Cn
of dimension d. If I(X) = (f1 , . . . , fs ), then, by Proposition 3.71,

∂fj
Xsing = {x ∈ X | rk (x) < n − d}
∂xi
which is the closed subset defined by the vanishing of all (n − d) × (n − d)
minors of the Jacobian matrix Jac(f1 , . . . , fs ). In order to see that Xsing has
a dense complement, we use the fact, that every irreducible variety contains
a special open set which is isomorphic to a special open set of an irreducible
hypersurface H (see Lemma 3.43). Since H contains a dense open set of
non-singular points (see Example 3.28 above) the claim follows.
Exercise 68. If X is an affine variety such that all irreducible components
have the same dimension, then Xsing is closed and has a dense complement.
(We will see later in Corollary 3.80 that this holds for every affine variety.)
Exercise 69. The hypersurface H ⊆ C3 from Exercise 43 is nonsingular.
Exercise 70. Let q ∈ C[x1, . . . , xn ] be a quadratic form and Q := V(q) ⊆ Cn .
Then 0 is a singular point of Q. It is the only singular point if and only if q is
nondegenerate.
Exercise 71. Determine the singular points of the plane curves

Ep := V(y 2 − p(x)),
where p(x) is an arbitrary polynomial, and deduce a necessary and sufficient con-
dition for Ep to be nonsingular.
Exercise 72. Let X ⊆ Cn be a closed cone (see Exercise 3). Then Xsing is
a cone, too. Moreover, 0 ∈ X is a nonsingular point if and only if X is a vector
subspace.
Exercise 73. Let X be an affine variety such that the group of automorphisms
acts transitively on X. Then X is smooth.
Associated graded algebras. Let R be a C-algebra and a ⊆ R an
ideal. The associated graded algebra gra R is defined in the following way.
Consider the C-vector space
M
gra R := ai /ai+1 = R/a ⊕ a/a2 ⊕ a2/a3 ⊕ · · ·
i≥0
and define the multiplication of (homogeneous) elements by

(f + ai+1 ) · (h + aj+1 ) := fh + ai+j+1
for f ∈ ai , h ∈ aj . It is easy to see that this defines a multiplication on gra R.
By definition, R/a is a subalgebra of gra R, and gra R is generated by a/a2 as
a R/a-algebra. In particular, if R is finitely generated as a C-algebra, then
so is gra R.
We want to use this construction to give the following characterization of
non-singular points.
Theorem 3.75. Let X be an affine variety. A point x ∈ X is nonsingular
if and only if the associated graded algebra grmx O(X) is a polynomial ring.
In particular, the local ring OX,x of a nonsingular point x is a domain and
so x belongs to a unique irreducible component of X.
Before we can give the proof we have to explain some technical results
from commutative algebra. Let R be a C-algebra and m ⊆ R a maximal
ideal. Consider the subalgebra R̃ of R[t, t−1] generated as an R-algebra by t
and mt−1 :
R̃ := R[t, mt−1] = · · · ⊕ m2 t−2 ⊕ mt−1 ⊕ R ⊕ Rt ⊕ Rt2 ⊕ · · · ⊆ R[t, t−1].
In the following lemma we collect some basic properties of this construction.
Lemma 3.76.
(1) If R is finitely generated then so is R̃.
∼
(2) There is a canonical isomorphism R̃/R̃t → grm R.
∼ g
(3) If a ⊆ m is an ideal and ã := a[t, t−1] ∩ R̃ then R̃/ã → R/a.
(4) If n ⊆ R is the nilradical, then ñ := n[t, t−1] ∩ R̃ is the nilradical of
∼ g
R̃, and R̃/ñ → R/n.
(5) Assume that R is a finitely generated domain. Then R̃ is a domain,
and we have
dim R̃ = dim R + 1 and dim R̃/R̃t = dim R.
(6) Assume that R is finitely generated and that the minimal primes
p1 , . . . , pr are all contained in m. Then the p̃1, . . . , p̃r are the minimal
primes of R̃.
Proof. (1) If R = C[h1, · · · , hm ] and m = (f1 , . . . , fn ) then
R̃ = C[h1, . . . , hm , t, f1t−1 , . . . , fn t−1 ],
and so R̃ is finitely generated.
(2) By definition, we have
R̃t = · · · ⊕ m3 t−2 ⊕ m2 t−1 ⊕ m ⊕ Rt ⊕ Rt2 ⊕ · · · .
Hence
R̃/R̃t = · · · ⊕ (m2 /m3 )t−2 ⊕ (m/m2 )t−1 ⊕ R/m
(3) The canonical map π : R[t, t−1] → (R/a)[t, t−1 ] induces, by our con-
struction, a surjective homomorphism π̃ : R̃ → R/a g with kernel ker π ∩ R̃ =
−1
a[t, t ] ∩ R̃.
(4) Put Rred := R/n. Then Rred [t, t−1] is reduced, i.e. without nilpotent el-
g
ements 6= 0, and so is R −1 −1
red. Since the kernel of the map R[t, t ] → Rred [t, t ]
−1
is equal to n[t, t ] and consists of nilpotent elements the claim follows from
(3).
(5) The first part is clear since R[t, t−1] is a domain. Since R̃t = R[t, t−1]
we get dim R̃ = dim R[t, t−1 ] = dim R[t] = dim R + 1. Moreover, by the
Principal Ideal Theorem (Theorem 3.54) we have dim R̃/R̃t = dim R̃ − 1.
T
(6) It follows from (3) and (5) that the ideals p̃i are prime. Since i pi = n
we obtain from (2)
\ \
p̃i = pi [t, t−1] ∩ R̃ = n[t, t−1] ∩ R̃ = ñ.
i i
Since p̃i ∩ R[t] = pi [t] there are no inclusions p̃i ⊆ p̃j for i 6= j, and the
claim follows. (We use here the well-known fact that the minimal T primes in
a finitely generated C-algebra are characterized by the condition pi = n,
cf. Remark 3.25.)
In the next lemma we give some properties of the associated graded al-
gebra grm R where m is a maximal ideal of R.
Lemma 3.77. Let R be a C-algebra and m ⊆ R a maximal ideal.
T
(1) Assume that j mj = (0). If grm R is a domain, then so is R.
(2) Denote by mRm ⊆ Rm the maximal ideal of the localization Rm .
∼
There is a natural isomorphism grm R → grmRm Rm of graded C-
algebras.
Proof. (1) If ab = 0 for non-zero elements a, b ∈ R, we can find i, j ≥ 0
such that a ∈ mi \mi+1 and b ∈ mj \mj+1 . Thus ā := a+mi+1 and b̄ := b+mj+1
are non-zero elements in grm A, and āb̄ = ab + mi+j+1 = 0. This contradiction
proves the claim.
(2) Set M := mRm ⊆ Rm . The image of S := R \ m in R/mk consists of
invertible elements and so R/mk → Rm /Mk is surjective. It is also injective,
because Rm /Mk can be identified with the localization of R/mk at S. Thus
∼ ∼
R/mk → Rm /Mk and so mi /mi+1 → Mi /Mi+1 for all i ≥ 0.
Finally, we need the following result due to Krull. It implies
T that in a
local Noetherian C-algebra R with maximal ideal m we have j≥0 mj = (0).
Lemma T 3.78 (Krull). Let R be a Noetherian C-algebra, a ⊆ R an ideal
and b := j≥0 aj . Then ab = b. In particular, there is an a ∈ a such that
(1 + a)b = 0.
Proof. The second claim follows from the first and the lemma of
Nakayama (Lemma 3.46). Let a = (a1, . . . , as ) and put I := hf | f ∈
R[x1 , . . . , xs ] homogeneous and f (a1, . . . , as ) ∈ bi. Note that I ⊆ R[x1 , . . . , xs ].
It is easy to see that I is an ideal of R[x1, . . . , xs ] and so I = (f1, . . . , fk )
where the fj are homogeneous. Choose an n ∈ N, n > deg fj for all j. By
definition, b ⊆ an and so, for every b ∈ b, there is a homogeneous polyno-
mial f ∈ R[x P 1 , · · · , xs ] of degree n such that f (a1, . . . , as ) = b. It follows
that f = j hj fj Pwhere the hj are homogeneous of degree > 0, and so
b = f (a1 , . . . , as ) = j hj (a1 , . . . , as )fj (a1, . . . , as ) ∈ ab.
The next proposition is a reformulation of our main Theorem 3.75. For
later use we will prove it in this slightly more general form.
Proposition 3.79. Let R be a finitely generated C-algebra and let m ⊆ R

be a maximal ideal. Then dim grm R = dim Rm . Moreover, dimC m/m2 =
dim Rm if and only if grm R is a polynomial ring. If this holds, then Rm is a
domain.
Proof. Inverting an element from R \ m does not change grm R (Lemma

3.77(2)). Therefore we can assume that all minimal primes of R are con-
tained in m. In particular, we have dim Rm = dim R = maxi dim R/pi where
p1 , . . . , pr are the minimal prime ideals. Moreover, every element from R \ m
is a non-zero divisor.
Now consider the C-algebra R̃ = R[t, mt−1] ⊆ R[t, t−1] introduced above.
It follows from Lemma 3.76 that R̃ has the following two properties:
∼
(i) R̃/R̃t → grm R, by (2).
(ii) dim R̃/R̃t = dim R, by (5) and (6).
Hence, dim grm R = dim Rm , proving the first claim.
Assume now that dimC m/m2 = dim Rm =: n. Then we obtain a surjective
homomorphism
ρ : C[y1, . . . , yn ] → grm R
by sending y1 , . . . , yn to a C-basis of m/m2 . But every proper residue class
ring of C[y1, . . . , yn ] has dimension < n, and so the homomorphism ρ is an
isomorphism.
On the other hand, if grm R is a polynomial ring then
dim Rm = dim grm R = dimC m/m2 .

T
Moreover, j>0 mj = (0) by Lemma 3.78, because every element from R \ m
is a non-zero divisor, and so R is a domain by Lemma 3.77(1).
Corollary 3.80. If X is an affine variety, then Xsing ⊆ X is a closed

subset whose complement is dense in X.
S
Proof. Let X = i Xi be the decomposition of X into irreducible com-
ponents. A point x ∈ Xi is a singular point of X if and only if it is either
a singular point of Xi or it belongs to two different irreducible components.
Thus
[ [
Xsing = (Xi )sing ∪ Xj ∩ Xk ,
i j6=k
and the claim follows easily.

Let us denote by ÔX,x the mx -adic completion of the local ring OX,x . It
is defined to be the inverse limit
ÔX,x := lim O(X)/mkx .
←−
(We refer to [26, I.7.1 and I.7.2] for more details and some basic properties.)
T
Since mkx = {0} we have a natural embedding OX,x ⊆ ÔX,x.
If X = Cn and x = 0 then the completion coincides with the algebra of
formal power series in n variables:
ÔCn ,0 = C[[x1, . . . , xn ]].
The next result is an easy consequence of Theorem 3.75 above.
Corollary 3.81. The point x ∈ X is non-singular if and only if ÔX,x
is isomorphic to the algebra of formal power series in dimx X variables.
Remark 3.82. A famous result of Auslander-Buchsbaum states that the
local ring OX,x in a nonsingular point of a variety X is always a unique
factorization domain. For a proof we refer to [55, §20, Theorem 20.3].
Vector fields S and tangent bundle. Let X be an affine variety. De-
note by T X := x∈X TxX the disjoint union of the tangent spaces and by
p : T X → X the natural projection, δ ∈ TxX 7→ x. We call T X the tangent
bundle of X. We will see later that T X has a natural structure of an affine
variety and that p is a morphism.
A section ξ : X → T X of p, i.e. p ◦ ξ = IdX or ξx := ξ(x) ∈ Tx X for all
x ∈ X, is a collection (ξx )x∈X of tangent vectors and thus can be considered
as an operator on regular functions f ∈ O(X):
ξf(x) := ξx f for x ∈ X.
Definition 3.83. An (algebraic) vector field on X is a section ξ : X →
T X with the property that ξf ∈ O(X) for all f ∈ O(X). The space of
algebraic vector fields is denoted by Vec(X).
(In the following, we will mostly talk about “vector fields” and omit the term
“algebraic” whenever it is clear from the context.)
Thus a vector field ξ can be considered as a linear map ξ : O(X) → O(X),
and so Vec(X) is a vector subspace of End(O(X)). More generally, the vector
fields form a module over O(X) where the product fξ for f ∈ O(X) is defined
in the obvious way: (f ξ)x := f (x)ξx .
Example 3.29. Let X = V be a C-vector space and fix a vector v ∈ V .

Then ∂v ∈ Vec(V ) is defined by x 7→ ∂v,x . It follows that

f (x + tv) − f (x)
∂v f := ∈ O(X)
t t=0
which implies that this vector field is indeed algebraic. We claim that every
algebraic vector field on V is of this form. In fact, if V = Cn then
M
n
∂
Vec(Cn ) = C[x1, . . . , xn ]
i=1
∂xi
which means that every algebraic vector field ξ on Cn is of the form ξ =

P ∂ n
i hi ∂xi where hi ∈ C[x1 , . . . , xn ] = O(C ). (This follows from the two facts
that every vector field ξ on Cn is of this form with arbitrary functions hi and
that ξ(xi ) = hi .)
Another observation is that for every vector field ξ on X the correspond-
ing linear map ξ : O(X) → O(X) is a derivation, i.e. ξ is a linear differential
operator:
ξ(f h) = f ξh + h ξf for all f, h ∈ O(X).
Proposition 3.84. The map sending a vector field to the corresponding
∼
linear differential operator defines a bijection Vec(X) → Der(O(X), O(X)) ⊆
End(O(X)).
Proof. It remains to show that every derivation ξ : O(X) → O(X) is
given by an algebraic vector field. For this, define ξx := evx ◦ξ. Then the
vector field (ξx )x∈X is algebraic and the corresponding linear map is ξ.
The Example 3.29 above shows that for X = V we have a canonical

isomorphism T X ' X × V , using the identifications Tx X = V ' {x} × V .
Then p : T X → X is identified with the projection prX and algebraic vector
fields correspond to the morphism ξ : X → X × V of the form ξ(x) = (x, ξx).
Proposition 3.85. Let X ⊆ V be a closed subset.
(1) If ξ ∈ Vec(V ) then ξ|X defines a vector field on X (i.e. ξx ∈ Tx X
for all x ∈ X) if and only if ξf = 0 for all f ∈ I(X). Moreover, it
suffices to test a system of generators of the ideal I(X).
∼
(2) There is a canonical bijection T X → {(x, δ) | δ ∈ TxX ⊆ V } where
the latter is a closed subset of X × V . Thus T X has the structure of
an affine variety. Using coordinates, we get

∼
Xn
∂f
T X → {(x, a1, . . . , an ) | ai (x) = 0 for all f ∈ I(X)} ⊆ X × Cn
i=1
∂x i
(3) A vector field ξ on X is algebraic if and only if ξ : X → T X is a

morphism.
Proof. (1) We have ξx ∈ Tx X for all x ∈ X if and only if ξx f = 0 for
all x and all f ∈ I(X) which is equivalent to ξf|X = 0 for all f ∈ I(X).
(2) We can assume that V = Cn and O(V ) = C[x1, . . . , xn ]. If I(X) =
(f1 , . . . fm ) then, by (1),
T 0X := {(x, δx) ∈ X × V | δ ∈ Tx X}
Xn
∂fj
= {(x, a1, . . . , an ) | ai (x) = 0 for j = 1, . . . , m} ⊆ X × Cn
i=1
∂x i
which shows that this is a closed subspace of X × Cn . Now (2) follows easily.
(3) Using the identification of T X with the closed
P subvariety T 0X above,
∂
an arbitrary section ξ : X → T X has the form ξx = hi (x) ∂x i
with arbitrary
functions hi on X. The vector field ξ is algebraic if and only if hi = ξx̄i is
regular on X which is equivalent to the condition that ξ : X → T X is a
morphism.
Example 3.30. Consider the curve H := V(xy − 1) ⊆ C2 . Then I(H) =
(xy − 1). For a vector field ξ = a(x, y)∂x + b(x, y)∂y on C2 we get
ξ(xy − 1) = a(x, y)y + b(x, y)x.
Thus ξ(xy −1)|H = 0 if and only if ay +bx = 0 on H. It follows that x∂x −y∂y
defines a vector field ξ0 on H and that Vec(H) = O(C)ξ0 . (In fact, setting
h := ay|H = −bx|H we get a|H = h · x|H and b|H = −h · y|H .)
The tangent bundle T H ⊆ H × C2 has the following description (see
Proposition 3.85(1)):
TH
= {(t, t−1, α, β) | αt−1 + βt = 0}
= {(t, t−1, −βt2, β | t ∈ C∗ , β ∈ C}
∼
→ H × C.
Example 3.31. Now consider Neil’s parabola C := V(y 2 − x3) ⊆ C2 (see
Example 3.8). Then a vector field a∂x + b∂y defines a vector field on C if and
only if
−3ax2 + 2by = 0 on C.
∼
To find the solutions we use the isomorphism O(C) → C[t2, t3], x 7→ t2, y 7→
t3 (see Example 3.19). Thus we have to solve the equation 3āt = 2b̄ in
C[t2, t3]. This is easy: Every solution is a linear combination (with coefficients
in C[t2, t3]) of the two solutions (2t2 , 3t3) and (2t3, 3t4 ). This shows that
ξ0 := (2x∂x + 3y∂y )|D and ξ1 := (2y∂x + 3x2∂y )|D
are vector fields on C and that Vec(C) = O(C)ξ0 + O(C)ξ1 . Moreover, x̄2ξ0 =
ȳξ1 .
Our calculation also shows that every vector field on C vanishes in the
singular point 0 of the curve. For the tangent bundle we get
T C = {(t2, t3, α, β) | −3αt4 + 2βt3 = 0} ⊆ C × C2
which has two irreducible components, namely
T C = {(t2, t3, 2α, 3αt) | t, α ∈ C} ∪ {(0, 0)} × C2
Exercise 74. Determine the vector fields on the curve D := V(y 2 − x2 − x3 ) ⊆
C2 . Do they all vanish in the singular point of D?
Exercise 75. Determine the vector fields on the curves D1 := {(t, t2, t3) ∈
C | t ∈ C} and D2 := {(t3 , t4, t5) ∈ C3 | t ∈ C}.
3
L
(Hint: For D2 one can use that O(D2) ' C[t3 , t4, t5] = C ⊕ i≥3 Cti .)
Proposition 3.86. The vector fields Vec(X) on X form a Lie algebra

with Lie bracket
[ξ, η] := ξ ◦ η − η ◦ ξ.
Proof. By Proposition 3.84 it suffices to show that for any two deriva-
tions ξ, η of O(X) the commutator ξ ◦ η − η ◦ ξ is again a derivation. But
this is a general fact and holds for any associative algebra, see the following
Exercise 77.
Exercise 76. Let A be an arbitrary associative C-algebra. Then A is a Lie
algebra with Lie bracket [a, b] := ab − ba, i.e., the bracket [ , ] satisfies the Jacobi
identity
[a, [b, c]] = [[a, b], c] + [b, [a, c]] for all a, b, c ∈ A.
Exercise 77. Let R be an associative C-algebra. If ξ, η : R → R are both C-
derivations, then so is the commutator ξ ◦ η − η ◦ ξ. This means that the derivations
Der(R) form a Lie subalgebra of EndC (R).
Exercise 78. Let X ⊆ Cn be a closed and irreducible. Then dim T X ≥
2 dim X. If X is smooth then T X is irreducible and smooth of dimension dim T X =
2 dim X.
(Hint: If I(X) = (f1 , . . . fm ) then T X ⊆ Cn × Cn is defined by the equations
n
X ∂fj
fj = 0 and yi (x) = 0 for j = 1, . . ., m.
∂xi
i=1
The Jacobian matrix of this system of 2m equations in 2n variables x1 , . . . , xn,

y1 , . . . , yn has the following block form

Jac(f1 , . . ., fm ) 0
∗ Jac(f1, . . . , fm )
and thus has rank ≥ 2 · rk Jac(f1, . . . , fm ) = 2(n − dim X).)
Differential of a morphism. Let ϕ : X → Y be a morphism of affine

varieties and let x ∈ X.
Definition 3.87. The differential of ϕ in x is the linear map
dϕx : TxX → Tϕ(x)Y
defined by δ 7→ dϕx (δ) := δ ◦ ϕ∗ .
If Z ⊆ X is a closed subvariety and z ∈ Z, then we get for the induced
morphism ϕ|Z : Z → Y that d(ϕ|Z )z = dϕz |Tz Z . Another obvious remark is
that the differential of a constant morphism is the zero map.
Remark 3.88. Set y := ϕ(x). The comorphism ϕ∗ : O(Y ) → O(X) de-
fines a homomorphism my → mx and thus a linear map ϕ̄∗ : my /m2y → mx /m2x .
It is easy to see that the differential dϕx corresponds to the dual map of ϕ̄∗
under the isomorphisms TxX ' Hom(mx /m2x , C) and Ty Y ' Hom(my /m2y , C)
(see Lemma 3.68).
Example 3.32. Using the identification T(x,y) X × Y = Tx X ⊕ Ty Y (see
Proposition 3.70) we easily see that the differential d(prX )x : T(x,y) X × Y →
Tx X coincides with the linear projection prTx X .
Proposition 3.89. Let ϕ = (f1 , . . . , fm ) : Cn → Cm , fj ∈ O(Cn ) =
C[x1, . . . , xn ]. Then the differential
dϕx : Tx Cn = Cn → Tϕ(x)Cm = Cm
of ϕ in x ∈ Cn is given by the Jacobi matrix

∂fj
Jac(f1 , . . . , fm )(x) = (x) .
∂xi i,j
Proof. The identification of the tangent space TxCn = Derx (O(Cn ))

with Cn is given by δ 7→ (δx1, . . . , δxn ) (see Example 3.25). This implies that
dϕx (δ) = ((δ ◦ ϕ∗ )(y1), . . . , (δ ◦ ϕ∗)(ym )) = (δf1, . . . , δfm).
Now the claim follows since
X
n
∂fj
δfj = (x) · δxi.
i=1
∂xi

Exercise 79. Let ϕ : X → Y and ψ : Y → Z be morphisms of affine varieties
and let x ∈ X. Then
d(ψ ◦ ϕ)x = dψy ◦ dϕx
where y := ϕ(x) ∈ Y .
In order to calculate explicitly differentials of morphisms we will again
use the algebra C[ε] of dual numbers (Remark 3.72). Recall that for δ ∈ Tx X
the map ρ := evx ⊕δε : O(X) → C[ε] is a homomorphism of algebras and
vice versa. If ϕ : X → Y is a morphism and x ∈ X, y := ϕ(x) ∈ Y , then we
obtain, by definition, the following commutative diagram:
evx ⊕δε
O(X) / C[ε]
O
oooo7
oo
ϕ∗
ooooo
oo yev ⊕dϕ x (δ)ε
O(Y )
If X := V and Y := W are vector spaces then a homomorphism ρ : O(V ) →
C[ε] corresponds to an element x ⊕ vε ∈ V ⊕ V ε where ρ(f ) = f (x + vε),
and so ρ ◦ ϕ∗ corresponds to the element ϕ(x + vε) ∈ W ⊕ W ε. Thus we
obtain the following result which is very useful for calculating differentials of
morphisms.
Lemma 3.90. Let ϕ : V → W be a morphism between vector spaces, and
let x ∈ V and v ∈ TxV = V . Then we have
ϕ(x + εv) = ϕ(x) + dϕx (v) ε
where both sides are considered as elements of W ⊕ W ε.
Example 3.33. The differential of the m-th power maps Mn → Mn ,
A 7→ Am , in E is m · Id. In fact, (E + Xε)m = E + mXε.
The differential of ϕ : M2 → M2 , ϕ(A) := A2, in an arbitrary matrix A is
given by dϕA (X) = AX + XA, because (A + Xε)2 = A2 + (AX + XA)ε.
The differential of the matrix multiplication µ : Mn × Mn → Mn in (E, E)

is the addition: (E + Xε)(E + Y ε) = E + (X + Y )ε.
Exercise 80. Consider the multiplication µ : M2 × M2 → M2 and show:
(1) dµ(A,B) is surjective, if A or B is invertible.
(2) If rk A = rk B = 1, then dµ(A,B) has rank 3.
(3) We have rk dµ(A,0) = rk dµ(0,A) = 2 rk A.
Exercise 81. Calculate the differential of the morphism ϕ : End(V ) × V → V

given by (ρ, v) 7→ ρ(v), and determine the pairs (ρ, v) where dϕ(ρ,v) is surjective.
Tangent spaces of fibers. Let ϕ : X → Y be a morphism, x ∈ X

and F := ϕ−1 (ϕ(x)) the fiber through x. Since ϕ|F is the constant map, its
differential in any point is zero and so Tx F ⊆ ker dϕx . This proves the first
part of the following result.
Proposition 3.91. Let ϕ : X → Y be a morphism, x ∈ X and F :=
−1
ϕ (ϕ(x)) the fiber through x.
(1) Tx F ⊆ ker dϕx .
(2) If the fiber F is reduced in x, then TxF = ker dϕx .
Proof. Put y := ϕ(x) ∈ Y . By definition the fiber is reduced in x if and
only if the ideal in the local ring OX,x generated by ϕ∗ (my ) is perfect, which
means that OF,x = OX,x /ϕ∗ (my )OX,x (see Definition 3.32).
Now let δ ∈ Tx X be a derivation of O(X) in x. If δ ∈ ker dϕx then
δ ◦ ϕ∗ = 0. Hence δ, regarded as a derivation of OX,x, vanishes on ϕ∗(my )OX,x
and thus induces a derivation of OF,x in x, i.e., δ ∈ TxF .
Example 3.34. Let X ⊆ Cn be a closed subset and I(X) = (f1 , . . . , fm ).
Consider the morphism ϕ = (f1 , . . . , fm ) : Cn → Cm . Then X = ϕ−1 (0), and
this fiber is reduced in every point. Thus, for every x ∈ X,
TxX = ker dϕx = ker Jac(f1 , . . . , fm )(x)
as we have already seen in Proposition 3.71.
Exercise 82. For every point (x, y) ∈ X × Y we have TxX = ker d(prY )(x,y)
and Ty X = ker d(prX )(x,y) where prX , prY are the canonical projections (see
Proposition 3.70).
Exercise 83. For the closed subset N ⊆ M2 of nilpotent 2 × 2-matrices we

have I(N ) = (tr, det).
Morphisms of maximal rank. The main result of this section is the

following theorem.
Theorem 3.92. Let ϕ : X → Y be a dominant morphism between two
irreducible varieties X and Y . Then there is a dense open set U ⊆ X such
that dϕx : TxX → Tϕ(x)Y is surjective for all x ∈ U .
We first work out an important example which will be used in the proof
of the proposition above.
Example 3.35. Let Y be an irreducible affine variety and XP⊆ Y × C
an irreducible hypersurface. Assume that I(X) = (f ) where f = ni=0 fi ti ∈
O(Y )[t] = O(Y × C) and fn = 1. Consider the following diagram:
⊆
X GG / Y ×C
GG
GG prY
p GGG
#
Y
∂f
Then the differential dp(y,a) : T(y,a)X → Ty Y is surjective if (y, a) 6= 0,
∂t
and this holds on a dense open set of X.
Proof. We have T(y,a) X ⊆ T(y,a) Y × C = Ty Y ⊕ C, and this subspace is

given by T(y,a)X = {(δ, λ) | (δ, λ)f = 0}, because I(X) = (f ). Now we have
X
n X
n
∂f
(δ, λ)f = (δfi · ai + fi (y) · i · ai−1 · λ) = δfi · ai + (y, a) · λ
i=0 i=0
∂t
Since dp(y,a) (δ, λ) = δ we see that dp(y,a) is surjective if ∂f

∂t
(y, a) 6= 0 which
∂f
proves the first claim. But ∂t cannot be a multiple of f and thus does not
vanish on X, proving the second claim.
The next lemma shows that the situation described in the example above
always holds on an open set for every morphism of finite degree.
Lemma 3.93. Let X, Y be irreducible affine varieties and ϕ : X → Y a
morphism of finite degree. Then there is a special open set U ⊆ Y and a
closed embedding γ : ϕ−1 (U ) ,→ U × C with the following properties:
P
(i) I(γ(U )) = (f ) where f = ni=0 fi ti ∈ O(U )[t];
(ii) prU ◦γ = ϕ|ϕ−1 (U ).
' ⊆ )
ϕ−1 (U ) / VU ×C (f ) /U ×C
MM r
MMM
MMM p rrrrr
ϕ MM rrr prU
M& yrrr
U
Proof. We have to show that there is a non-zero s ∈ O(Y ) such that
O(X)s ' O(Y )s [t]/(f ) with a polynomial f ∈ O(Y )s [t]. Then the claim
follows by setting U := Ys .
By assumption, the field C(X) is a finite extension of C(Y ) of degree n,
say,
C(X) = C(Y )[h] ' C(Y )[t]/(f )
Pn i
where f = i=0 fi t , fi ∈ C(Y ) and fn = 1. There is a non-zero element
s ∈ O(Y ) such that
(a) fi ∈ O(Y )s for all i,
(b) h ∈ O(X)s and L
(c) O(X)s = O(Y )s [h] = n−1 i
i=0 O(Y )s h .
In fact, (a) and (b) are clear. For (c) we first remark that O(Y )s [h] =
Ln−1 i
i=0 O(Y )s h ⊆ O(X)s , because of (a) and (b). If h1 , . . . , hm is a set of gen-
erators of O(X) we can find a non-zero s ∈ O(Y ) such that hi ∈ O(Y )s [h],
proving (c).
Setting U := Ys we get ϕ−1 (U ) = Xs and O(Xs ) = O(Ys )[t]/(f ), by (c),
Proof of Theorem 3.92. By the Decomposition Theorem (Theorem
3.59) we can assume that ϕ is the composition of a finite surjective morphism
and a projection of the form Y × Cr → Y . Since the differential of the second
morphism is surjective in any point we are reduced to the case of a finite
morphism. Now the claim follows from Lemma 3.93 above and Example 3.35.

Lemma 3.94. Let ϕ : X → Y be a morphism, x ∈ X and y := ϕ(x) ∈ Y .
Assume that X is smooth in x and dϕx is surjective. Then,
(1) Y is smooth in y;
(2) The fiber ϕ−1 (y) is reduced and smooth in x, and dimx F = dimx X −
dimy Y .
Proof. By assumption,
dim TxF ≤ dim ker dϕx = dim TxX − dim Ty Y ≤ dim X − dim Y ≤ dimx F
which implies that we have equality everywhere. In particular, X and F are

both smooth in x.
If we denote by m̄ ⊆ O(X)/my O(X) the maximal ideal corresponding
to x ∈ F one easily sees that m̄/m̄2 is the cokernel of the natural map
my /m2y → mx /m2x induced by ϕ∗ . The duality between mx /m2x and TxX (see
Lemma 3.68 and Remark 3.88) implies that dim ker dϕx = dimC m̄/m̄2 . Since
dim ker dϕx = dimx F = dim O(X)x /my O(X)x it follows that O(X)x /my O(X)x
is a domain (Proposition 3.79), and so F is reduced in x.
Corollary 3.95. For every morphism ϕ : X → Y there is a dense spe-
cial open set U ⊆ X such that all fibers of the morphism ϕ|U : U → Y are
reduced and smooth.
Proof. One easily reduces to the case where X is irreducible. Then there
is a special open set U ⊆ X which is smooth (Corollary 3.80) and such that
dϕx is surjective for all x ∈ U (Theorem 3.92). Now the claim follows from
the previous Lemma 3.94.
Corollary 3.96 (Lemma of Sard). Let ϕ : Cn → Cm be a dominant
morphism and set S := {x ∈ Cn | dϕx is not surjective}. Then S is closed
and ϕ(S) is a proper closed subset of Cm . In particular, there is a dense open
set U ⊆ Cm such that all fibers ϕ−1 (y) for y ∈ U are reduced and smooth of
dimension n − m.
Proof. If ϕ = (f1, . . . , fm ) then
S = {x ∈ Cn | rk Jac(f1, . . . , fm )(x) < m}
and so S is closed in Cn . Moreover, the differential of ϕ|S : S → Cm at any
point of S is not surjective. Therefore, by Theorem 3.92, the closure of the
image ϕ(S) has dimension strictly less than m.
Exercise 84. Let f ∈ C[x1, . . . , xn ] be a non-constant polynomial. Then
V(f − λ) is a smooth hypersurface for almost all λ ∈ C.
Corollary 3.97. If ϕ : X → Y is a morphism such that dϕx = 0 for all

x ∈ X, then the image ϕ(X) is finite. In particular, if X is connected then
ϕ is constant.
Proof. If X 0 ⊆ X is an irreducible component and Y 0 := ϕ(X 0 ), then
the induced morphism ϕ0 : X 0 → Y 0 has the same property, namely dϕ0x = 0
for all x ∈ X 0 . It follows now from Theorem 3.92 that dim Y 0 = 0. Hence ϕ
is constant on X 0 .
5. NORMAL VARIETIES AND DIVISORS 101
Example 3.36. Let V be a vector space and W ⊆ V a subspace. If

X ⊆ V is a closed irreducible subvariety such that TxX ⊆ W for all x ∈ X
then X ⊆ x + W for any x ∈ X.
(This follows from the previous corollary applied to the morphism ϕ : X →
V/W induced by the linear projection V → V/W .)
5. Normal Varieties and Divisors
Normality.
Definition 3.98. Let A ⊆ B be rings. An element b ∈ B is integral over
A if b satisfies an equation of the form
X
n−1
n
b = ai bi where ai ∈ A.
i=0
Equivalently, b ∈ B is integral over A if and only if the subring A[b] ⊆ B is

a finite A-module.
If every element from B is integral over A we say that B is integral over
A.
Exercise 85. Let A ⊆ B be rings. If A is Noetherian and B finite over A,
then B is integral over A.
Lemma 3.99. Let A ⊆ B ⊆ C be rings and assume that A is Noetherian.
(1) If B is integral over A and C integral over B, then C is integral
over A.
(2) The set
B 0 := {b ∈ B | b is integral over A}
is a subring of B.
P
Proof. (1) Let c ∈ C. Then we have an equation cm = m−1 j
j=0 bj c with
bj ∈ B. In particular, the coefficients bj are integral over A and so, by in-
duction, A1 := A[b0, b1 , . . . , bm−1 ] is a finitely generated A-module. Moreover,
A1[c] is a finitely generated A1 -module, hence a finitely generated A-module.
But then A[c] ⊆ A1[c] is also finitely generated.
(2) Let b1 , b2 ∈ B 0. Then A[b1] is integral over A and b2 is integral over
A, hence integral over A[b1], and so A[b1, b2] is integral over A[b1]. Thus, by
(1), A[b1, b2] is integral over A which implies that b1 + b2 and b1b2 are both
integral over A, hence belong to B 0.
Exercise 86. Let f ∈ C[x] be a non-constant polynomial. Then C[x] is integral
over the subalgebra C[f ].
Definition 3.100. Let A be a domain with field of fractions K. We call

A integrally closed if the following holds:
If x ∈ K is integral over A then x ∈ A.
An affine variety X is normal if X is irreducible and O(X) is integrally closed.
We say that X is normal in x ∈ X if the local ring OX,x is integrally closed.
Example 3.37. A unique factorization domain A is integrally closed. In
particular, Cn is a normal variety.
(Let
Pn−1K be the field of fractions of A and x ∈ K integral over A: xn =
i
i=0 ai x where ai ∈P A. Write x = ab where a, b ∈ A have no common
divisor. Then an = b( n−1i=0 ai b
n−i−1 i
a ) which implies that b is a unit in A
and so x ∈ A.)
Exercise 87. If the domain A is integrally closed, then so is every ring of
fractions AS where 1 ∈ S ⊆ A is multiplicatively closed.
Lemma 3.101. Let X be an irreducible variety. Then X is normal if and
only if all local rings OX,x are integrally closed.
Proof. If X is normal then OX,x = O(X)mx is integrally closedT (see the
Exercise above), and the reverse implication follows from O(X) = x∈X OX,x
(Exercise 26).
Integral closure and normalization.
Proposition 3.102. Let A be a finitely generated C-algebra with no zero-
divisors 6= 0 and with field of fractions K, and let L/K be a finite field
extension. Then
A0 := {x ∈ L | x is integral over A} ⊇ A
is a finitely generated C-algebra which is finite over A.
Proof. We already know that A0 is a C-algebra (Lemma 3.99(2)).
(a) We first assume that A = C[z1, . . . , zm] is a polynomial ring and
K = C(z1, . . . , zm ). Let L = K[x] where x is integral over A and [L : K] =: n.
Denote by x1 := x, x2, . . . , xn the conjugates of x in some Galois extension
L0 of K. Clearly, all xj are integral over A, because they satisfy the same
equation asP x.
n−1 i
If y = i=0 ci x (ci ∈ K) is an arbitrary element of L we obtain the
“conjugates” of y in L0 in the form
X
n−1
yj = ci xij for j = 1, . . . , n.
i=0
Q
The n × n-matrix X := (xij ) has determinant d = j<k (xj − xk ) which is
integral over A. Obviously, d2 is symmetric, hence fixed under the Galois
group of L0/K, and so d2 ∈ K. Since d2 is also integral over A we finally get
d2 ∈ A. From Cramer’s rule we obtain
     
c1 y1 y1
. −1 . 1
 ..  = X  ..  = Adj(X)  ... 
d
cn yn yn
This shows that if y is integral over
P A then so is dci for all i, hence d2 ci ∈ A
n−1
for all i. This implies that d2 A0 ⊆ i=0 Axi , and so A0 is a finitely generated
A-module.
(b) For the general case we use Noether’s Normalization Lemma (Theo-
rem 3.48) which states that A contains a polynomial ring A0 = C[x1, . . . , xm]
such that A is finite over A0 . Thus A is integral over A0 and therefore, by
Lemma 3.99(1)
A0 = {x ∈ L | x is integral over A0}.
It follows from part (a) that A0 is a finitely generated A0-module, hence also
a finitely generated A-module.
Definition 3.103. Let A be a finitely generated C-algebra with no zero-
divisors 6= 0. If L is a finite field extension of the field of fractions of A,
then
A0 := {x ∈ L | x is integral over A} ⊇ A
is called the integral closure of A in L. Clearly, A0 is integrally closed.
Let X be an irreducible affine variety and denote by O(X)0 ⊆ C(X) the
integral closure of O(X) in its field of fractions C(X). By Proposition 3.102
there is a normal variety X̃ and a finite birational morphism η : X̃ → X such
that O(X̃) ' O(X)0 . More precisely, we have the following result.
Lemma 3.104. Let X be an irreducible variety and η : X̃ → X a mor-
phism with the following two properties:
(1) X̃ is normal;
(2) η is finite and birational.
Then O(X̃) is the integral closure of η ∗ (O(X)) in C(X̃) = η ∗ (C(X)), and we
have the following universal property:
(P) If Y is a normal affine variety then every dominant morphism ϕ : Y →
X factors through η: There is a uniquely determined ϕ̃ : Y → X̃ such
that ϕ = η ◦ ϕ̃:
8 X̃
ϕ̃q qq
η
q qq
q ϕ
Y /X
Proof. Since η is birational we have η ∗(O(X)) ⊆ O(X̃ ) ⊆ C(X̃) =

η ∗ (C(X)). By (2) O(X̃ ) is finite, hence integral over η ∗(O(X)), and by (1) it
is the integral closure of η ∗ (O(X)).
If Y is a normal affine variety and ϕ : Y → X a dominant morphism then
∼
O(X) → ϕ∗(O(X)) ⊆ O(Y ) ⊆ C(Y ).
Denote by O(X)0 the integral closure of O(X) in C(X). Since O(Y ) is inte-
grally closed it follows that ϕ∗ (O(X)0 ) ⊆ C(Y ) is contained in O(Y ). Since
∼
η ∗ induces an isomorphism O(X)0 → O(X̃) there is a uniquely determined
homomorphism ρ : O(X̃ ) → O(Y ) which makes the following diagram com-
mutative:
O(X̃)
O Z
~~~
~~ '
ρ ~~
~~
~~
~ O(X)0 η∗
~ n O
~~~ nnnnn
~~ nnn ⊆
~~~wnnnn ϕ∗ ?
O(Y ) o O(X)
Clearly, the corresponding morphism ϕ̃ : Y → X̃ is the unique morphism

such that ϕ = η ◦ ϕ̃.
Definition 3.105. The morphism η : X̃ → X constructed above is called
the normalization of X. It follows from Lemma 3.104 that it is unique up to
a uniquely determined isomorphism.
Exercise 88. If ϕ : X → Y is a finite surjective morphism where X is irre-
ducible and Y is normal, then #ϕ−1(y) ≤ deg ϕ for all y ∈ Y . (See Proposition 3.65
and its proof.)
Proposition 3.106. Let X be an irreducible variety. Then the set

Xnorm := {x ∈ X | X is normal in x}
is open and dense in X.
Proof. Let O(X)0 ⊆ C(X) be the integral closure of O(X) and define
a := {f ∈ O(X) | f O(X)0 ⊆ O(X)}.
Then a is a non-zero ideal of O(X), because O(X)0 is finite over O(X), and
Xnorm = X \ VX (a). In fact, for S := O(X) \ mx we have
OX,x = O(X)S ⊆ O(X)0S
and the latter is the integral closure of OX,x . On the other hand, O(X)S =
O(X)0S if and only if S ∩ a 6= ∅ which is equivalent to x ∈
/ VX (a).
Exercise 89. Consider the morphism ϕ : C2 → C4 , (x, y) 7→ (x, xy, y 2, y 3).
(1) ϕ is finite and ϕ : C2 → Y := ϕ(C2) is the normalization.
(2) 0 ∈ Y is the only non-normal and the only singular point of Y .
(3) Find defining equations for Y ⊆ C4 and generators of the ideal I(Y ).
Exercise 90. If X is a normal variety then so is X × Cn .
Discrete valuation rings and smoothness. Let K be a field.

Definition 3.107. A discrete valuation of the field K is a surjective map
ν : K ∗ := K \ {0} → Z with the following properties:
(a) ν(xy) = ν(x) + ν(y);
(b) ν(x + y) ≥ min(ν(x), ν(y)).
To simplify the notation one usually defines ν(0) := ∞.
Example 3.38. Let K = Q and p ∈ N be a prime number. Define
νp (x) := r ∈ Z if p occurs with exponent r in the rational number x 6= 0.
Then νp : Q∗ → Z is a discrete valuation of Q.
The following lemma collects some facts about discrete valuations. The
easy proofs are left to the reader.
Lemma 3.108. Let K be a field and ν : K ∗ → Z a discrete valuation.
(1) A := {x ∈ K | ν(x) ≥ 0} is a subring of K.
(2) m := {x ∈ K | ν(x) > 0} ⊆ A is a maximal ideal of A.
(3) {x ∈ K | ν(x) = 0} are the units of A.
(4) For every non-zero x ∈ K we have x ∈ A or x−1 ∈ A.
(5) m = (x) for every x ∈ K with ν(x) = 1.
(6) mk = {x ∈ K | ν(x) ≥ k} and these are all non-zero ideals of A.
(7) If m = (x) then every z ∈ K has a unique expression of the form
z = txk where k ∈ Z and t is a unit of A.
Definition 3.109. A domain A is called a discrete valuation ring if there

is a discrete valuation ν of its field of fractions K such that A = {x ∈ K |
ν(x) ≥ 0}. In particular, A has all the properties listed in Lemma 3.108
above.
Exercise 91. Let A be a discrete valuation ring with field of fractions K. If
B ⊆ K is a subring containing A then either B = A or B = K.
In the sequel we will use the following characterization of a discrete val-
uation ring (see [7, Proposition 9.2]).
Proposition 3.110. Let A be a Noetherian local domain of dimension 1,
i.e. the maximal ideal m 6= (0) and (0) are the only prime ideals in A. Then
the following statements are equivalent:
(i) A is a discrete valuation ring.
(ii) A is integrally closed.
(iii) The maximal ideal m is principal.
(iv) dimA/m m/m2 = 1.
(v) Every non-zero ideal of A is a power of m.
(vi) There is an x ∈ A such that every non-zero ideal of A is of the form
(xk ).
Proof. (i)⇒(ii): If x ∈ K and x ∈ / A then A[x] = K which is not finite
over A.
(ii)⇒(iii): Let a ∈ m, a 6= 0. Then mk ⊆ (a) and mk−1 6⊆ (a) for some
k > 0. Choose an element b ∈ mk−1 \ (a) and put x := ab . Then x−1m =
1
a
bm ⊆ a1 mk ⊆ A. If x−1 m ⊆ m then x−1 would be integral over A and so
x ∈ A, contradicting the construction. Thus x−1 m = A and so m = (x).
−1
(iii)⇒(iv): If m = (x) then m/m2 = A/m · (x + m2 ), and m2 6= m.

√
(iv)⇒(v): Let a ⊆ A be a non-zero ideal. Then a = m and so mk ⊆ a
for some k ∈ N. Put Ā := A/mk and denote by m̄ ⊆ Ā the image of m. Since
m = (x) + m2 we get m = (x) + mk for all k ∈ N and so m̄ = (x̄) ⊆ Ā. Now
it is easy to see that ā = m̄r for some r ≤ k, and so a = mr .
(v)⇒(vi): We have m 6= m2 . Choose x ∈ m \ m2 . Then, by assumption,
(x) = mk for some k ≥ 1, and so m = (x).
(vi)⇒(i): By assumption, every element a ∈ A has a unique expression of
the form a = txk where k ∈ N and t a unit of A. Define ν(a) := k. This has
a well-defined extension to K ∗ by setting ν( ab ) := ν(a) − ν(b) for a, b ∈ A,
b 6= 0. One easily verifies that ν is a discrete valuation of K and that A is
the corresponding valuation ring.
If Y is an irreducible curve, i.e. dim Y = 1, then the local rings OY,y

satisfy the assumptions of the proposition above. The equivalence of (i), (ii)
and (iv) then gives the following result. (In fact, we do not need to assume
that Y is irreducible; cf. Theorem 3.75.)
Proposition 3.111. Let Y be an affine variety and y ∈ Y such that
dimy Y = 1. Then the following statements are equivalent:
(i) The local ring OY,y is a discrete valuation ring.
(ii) Y is normal in y.
(iii) Y is smooth in y.
In particular, a normal curve is smooth and an irreducible smooth curve is
normal.
Remark 3.112. Let X be an irreducible variety and H ⊆ X an irre-
ducible hypersurface, i.e. codimX H = 1. The ideal p := I(H) of H is a min-
imal prime ideal 6= (0) and thus the localization OX,H := O(X)p is a local
Noetherian domain of dimension 1. If X is normal it follows from Propo-
sition 3.110 that OX,H is a discrete valuation ring which corresponds to a
discrete valuation νH : C(X)∗ → Z.
E.g. if f ∈ C[x1, . . . , xn ] is a non-constant irreducible polynomial and
H := V(f ), then the valuation νH has the following description: For a rational
function r ∈ C(x1, . . . , xn ) we have νH (r) = m if f occurs with exponent m
in a primary decomposition of r.
Normal varieties. We start with the following generalization of the
previous result saying that normal curves are smooth (Proposition 3.111).
Recall that the singular points Xsing of an affine variety form a closed subset
with a dense complement (Proposition 3.80).
Proposition 3.113. Let X be a normal affine variety. Then
codimX Xsing ≥ 2.
Proof. (a) Let H ⊆ X be an irreducible hypersurface and assume that
I(H) = (f ). We claim that if x ∈ H is a singular point of X then x is a
singular point of H, too. In fact, O(H) = O(X)/(f ) and mH,x = mx /fO(X).
Thus mH,x/m2H,x = (mx /m2x )/C · f¯ and so dim Tx H ≥ dim TxX − 1 > dim X −
1 = dim H.
(b) Now assume that codimX Xsing = 1, and let H ⊆ Xsing be an irre-
ducible hypersurface of X. If p := I(H) is a principal ideal it follows from
(a) that H consists of singular points. But this contradicts the fact that the
smooth points of an irreducible variety form a dense open set.
In general, the localization OX,H is a discrete valuation ring (Remark 3.112)

and therefore its maximal ideal pOX,H is principal (Proposition 3.110). This
implies that we can find an element s ∈ O(X) \ p such that the ideal
pO(X)s ⊆ O(X)s = O(Xs ) is principal. Since pO(X)s = I(H ∩ Xs ) we arrive
again at a contradiction, namely that all points of H ∩ Xs are singular.
Another important property of normal varieties is that regular functions
can be extended over closed subset of codimension ≥ 2.
Proposition 3.114. Let X be a normal affine variety and f ∈ C(X) a
rational function which is defined on an open set U ⊆ X. If codimX X \U ≥ 2
then f is a regular function on X.
Proof. Define the ”ideal of denominators”
a := {q ∈ O(X) | q · f ∈ O(X)}.
By definition U ⊆ V \ VX (a) and so, by assumption, codimX VX (a) ≥ 2.
Using Noether’s Normalization Lemma (Theorem 3.48) we can find
a finite surjective morphism ϕ : X → Cn . We have ϕ(VX (a)) = V(a ∩
C[x1, . . . , xn ]) and dim ϕ(VX (a)) = dim V(a ∩ C[x1, . . . , xn ]) ≤ n − 2. This
implies that we can find two polynomials q1, q2 ∈ a ∩ C[x1, . . . , xn ] with no
common divisor (see the following Exercise 92). As a consequence, we have
p1 p2
f= = for suitable p1 , p2 ∈ O(X).
q1 q2
If f (1) := f, f (2), . . . , f (d) are the conjugates of f in some finite field ex-
tension L/C(x1 , . . . , xn ) of degree d containing C(X) we have
pi1 pi
f (i) = = 2 for i = 1, . . . , d
q1 q2
(i) (i)
where the p1 are the conjugates of p1 and the p2 the conjugates of p2 . The
element f ∈ C(X) satisfies the equation
Y
d X
d
(i) d
(t − f ) = t + bj tn−j = 0
i=1 j=1
where the coefficients bj ∈ C(x1 , . . . , xn ) are given by the elementary sym-

metric functions σj in the following form:
1 (1) (d) 1 (1) (d)
bj = ±σj (f (1) , . . . , f (d) ) = ± j σj (p1 , . . . , p1 ) = ± j σj (p2 , . . . , p2 ).
q1 q2
Since p1 , p2 ∈ O(X) are integral over C[x1, . . . , xn ] we see that both
(1) (d) (1) (d)
σj (p1 , . . . , p1 ) and σj (p2 , . . . , p2 ) belong to C[x1, . . . , xn ]. Since q1 and q2
have no common factor this implies that bj ∈ C[x1, . . . , xn ]. As a consequence,

f is integral over C[x1, . . . , xn ] and thus belongs to O(X).
Exercise 92. Let a ⊆ C[x1, . . . , xn ] be an ideal with the property that any
two elements f1 , f2 ∈ a have a non-trivial common divisor. Then there is a non-
constant h which divides every element of a.
T
Corollary 3.115. If X is a normal variety then O(X) = p O(X)p
where p runs through the minimal prime ideals 6= (0).
T
Proof. Let r ∈ p O(X)p and define a := {q ∈ O(X) | q · r ∈ O(X)}.
It follows that a 6⊆ p for all minimal primes p 6= 0, and so VX (a) does not
contain an irreducible hypersurface. This implies that codimX VX (a) ≥ 2 and
so r is regular by the Proposition 3.114 above.
We thus have the following characterization of normal varieties. An irre-
ducible variety X is normal if and only if the following two conditions hold:
(a) For every minimal prime p 6= (0) the local ring O(X)p is a discrete
valuationTring;
(b) O(X) = p O(X)p where p runs through the minimal prime ideals
6= (0).
We have seen in examples that there are bijective morphisms which are
not isomorphisms. This cannot happen if the target variety is normal.
Proposition 3.116. Let X be an irreducible and Y a normal affine va-
riety and let ϕ : X → Y be a dominant morphism. Assume
(a) codimY Y \ ϕ(X) ≥ 2, and
(b) deg ϕ = 1.
Then ϕ is an isomorphism.
Proof. By assumption (b), we have the following commutative diagram:
⊆
O(Y ) −−−→ O(X)
 

⊆y

⊆y
C(Y ) C(X)
If H ⊆ Y is an irreducible hypersurface then, by assumption (a), H meets
the image ϕ(X) in a dense set and so ϕ(ϕ−1 (H)) = H. This implies that
there is an irreducible hypersurface H 0 ⊆ X such that ϕ(H 0 ) = H. If we
denote by p := I(H) ⊆ O(Y ) and p0 := I(H 0) ⊆ O(X), the corresponding
minimal prime ideals, we get p0 ∩ O(Y ) = p. Thus
O(Y )p ⊆ O(X)p0 $ C(Y ) = C(X).
Since O(Y )p is a discrete valuation ring this implies O(Y )p = O(X)p0 (see
Exercise 91). Thus, by Corollary 3.115,
\ \
O(X) ⊆ O(X)p0 = O(Y )p = O(Y ),
p0 p
There is a partial converse of Proposition 3.113 which is a special case

of the so-called Serre Criterion for Normality which we will explain below
without giving a proof.
Proposition 3.117. Let H ⊆ Cn be an irreducible hypersurface. If the
singular points Hsing have codimension ≥ 2 in H, then H is normal.
Example 3.39. Let Qn := V(x21+x22 +· · ·+x2n ) ⊆ Cn . Then dim Qn = n−1
and 0 ∈ Qn is the only singular point. Thus Qn is normal for n ≥ 3.
Exercise 93. Show that the nilpotent cone N := {A ∈ M2 | A nilpotent} is
a normal variety.
Proposition 3.118 (Serre’s Criterion). Let X ⊆ Cn be the zero set of

f1 , . . . , fr ∈ C[x1, . . . , xn ]: X := V(f1 , . . . , fr ). Define
X 0 := {x ∈ X | rk Jac(f1 , . . . , fr )(x) < r}.
(1) If X \ X 0 is dense in X then I(X) = (f1 , . . . , fr ) and X 0 = Xsing .
(2) If codimX X \ X 0 ≥ 2 then X is normal.
Example 3.40. Let N := {A ∈ Mn | A nilpotent} be the nilpotent cone
in Mn . We claim that N is a normal variety.
Proof. Consider the morphism π : Mn → Cn ,

π(A) := (tr A, tr A2, . . . , tr An ).
Then N = π −1(0). If P ∈ N is a nilpotent element of rank n − 1 then
rk dπP = n. In fact, tr(P + εX)k = tr(P k + εkP k−1 X) = εk tr(P k−1 X).
Taking P in Jordan normal form one easily sees that
dπP : X 7→ (tr X, tr P X, tr P 2 X, . . . , tr P n−1 X)
is surjective. It follows that rk Jac(f1 , . . . , fn )(P ) = n for the functions
fj (A) := tr Aj and for P ∈ N 0 := {nilpotent matrices of rank n − 1}. Now
one shows that codimN N \ N 0 = 2.
Divisors. Let X be a normal affine variety. Define

H := {H ⊆ X | H irreducible hypersurface}.
Definition 3.119. A divisor on X is a finite formal linear combination
X
D= nH · H where nH ∈ Z.
H∈H
We write D ≥ 0 if nH ≥ 0 for all H ∈ H. The set of divisors forms the

divisor group M
Div X = Z · H.
H∈H
Recall that for any irreducible hypersurface H ∈ H we have defined a

discrete valuation νH : C(X)∗ → Z whose discrete valuation ring is the local
ring OX,H (see Remark 3.112).
Definition 3.120. For f ∈ C(X)∗ we define the divisor of (f ) by
X
(f ) := νH (f ) · H.
H∈H
Such a divisor is called a principal divisor.

Remarks 3.121. (1) (f ) is indeed a divisor, i.e. νH (f ) 6= 0 only for
finitely many H ∈ H.
(This is clear for f ∈ O(X) \ {0}, because νH (f ) > 0 if and only if
H ⊆ V(f ), and follows for a general f = pq because (f ) = (p) − (q),
by definition.)
(2) (f · h) = (f ) + (h) for all f, h ∈ C(X).
(3) (f ) ≥ 0 if and only if f ∈ O(X). T
(We have νH (f ) ≥ 0 if and only if f ∈ OX,H . Since H∈H OX,H =
O(X) the claim follows.)
(4) (f ) = 0 if and only if f is a unit in O(X).
(If (f ) = 0 then, by (3), f ∈ O(X) and f −1 ∈ O(X).)
Definition 3.122. Two divisors D, D0 ∈ Div X are called linearly equiv-
alent , written D ∼ D0 , if D − D0 is a principal divisor. The set of equivalence
classes is the divisor class group of X:
Cl X := Div X/{principal divisors}
It follows that we have an exact sequence of commutative groups
1 → O(X)∗ → C(X)∗ → Div X → Cl X → 0
Remark 3.123. We have Cl X = 0 if and only if O(X) is a unique

factorization domain. In fact, a unique factorization domain is characterized
by the condition that all minimal prime ideals p 6= (0) are principal.
Example 3.41. Let C ⊆ C2 be a smooth curve. If f ∈ O(C) and f˜ ∈
C[x, y] is a representative of f , then
X
(f ) = mP · P,
˜
P ∈C∩V(f)
and the integers mP > 0 can be understood as the intersection multiplicity

of C and V(f˜) in P . E.g. if the intersection is transversal, i.e.
TP C ∩ TP V(f˜) = (0),
then mP = 1 (see the following Exercise 94).
Exercise 94. Let C, E ⊆ C2 be two irreducible curves, I(C) = (f ) and
I(E) = (h). If P ∈ C ∩ E define mP := dimC C[x, y]/(f, h). Show that
P
(1) If C is smooth and h̄ = h|C ∈ O(C), then (h̄) = P ∈C∩E mP · P
(2) If P ∈ C ∩ E and TP C ∩ TP E = (0) then mP = 1.
Exercise 95. (1) For the parabola C = V(y − x2) we have Cl C = (0).
(2) For an elliptic curve E = V(y 2 − x(x2 − 1)) every divisor D is linearly
equivalent to 0 or to P for a suitable point P ∈ E.
CHAPTER 4
Embedding problems in affine algebraic geometry
Peter Russell
1. The problem and first examples

Definition 4.1. Let X, Y be algebraic varieties. Subvarieties
Y1 ⊆ X, Y2 ⊆ X, Yi ' Y, i = 1, 2,
define equivalent embeddings of Y in X if there exists an automorphism ϕ
of X such that ϕ(Y1 ) = Y2 . We write Y1 ∼ Y2 .
Remark 4.2. A more stringent condition is that every isomorphism ψ :
Y1 → Y2 extends to an automorphism of X.
Example 4.1. Let X = A2 = Spec k[x, y].
(i) Let Y1 = V (y), Y2 = V (2x3 + x4 + (−1 + 2x + 2x2 )y + y 2). Then
Yi ' A1 and Y1 ∼ Y2 (exercise!).
(ii) Let char(k) = 2 and Y1 = V (y), Y2 = V (y 4 + x + x6). Then Y2 ' Y1 '
A1 and Y1 Y2 (exercise!). For more examples of this type see [1, 2].
(iii) Let Y1 = V (xy − 1), Y2 = V (xy 2 − 1). Then Y1 ' Y2 ' A1 \ {0} =
A1∗ and Y1 Y2 . (exercise!). In fact if Ya,b = V (xa y b − 1), 1 ≤ a ≤
b, GCD(a, b) = 1, thenYa,b ' A1∗ and Ya,b ∼ Ya0 ,b0 ⇐⇒ a = a0, b = b0.
(This does not exhaust all equivalence classes of embeddings of A1∗ in A2 .)
Exercise 96. Show that ψ : x → 1/x, y → 1/y defines an automorphism
of Ya,b . Show that ψ extends to an automorphism of A2 ⇐⇒ a = b = 1.
Most of the time we will consider closed embeddings (Y1 , Y2 closed in X),
but open embeddings (Y1 , Y2 open in X) can be interesting as well.
Example 4.2. All open embeddings of Y = A2 in X = P2 are equivalent
to the obvious one, the complement of a projective line. This follows from
the following:
Proof. Let F (X0 , X1 , X2 ) be a homogeneous polynomial of positive de-
gree d. Then X = P2 \ V+ (F ) is an affine variety. More precisely, A = Γ(X)=
113
114 4. EMBEDDING PROBLEMS IN AFFINE ALGEBRAIC GEOMETRY
rational functions on P2 with poles only on V+ (F ) is finitely generated over

k and X ' Spec A: the Veronese map given by the monomials of degree d
defines an isomorphism of P2 with a closed subvariety X̃ of a higher dimen-
sional projective space P and X is isomorphic to the complement in X̃ of a
hyperplane in P.
We suppose F is reduced.
a) Suppose F is irreducible. Then A is factorial ⇐⇒ d = 1 (the divisor
class group is cyclic of degree d).
b) If F is reducible, then A∗ 6= k ∗ .
We note that in Example 4.2 only linear automorphisms of A2 extends
to P2 .
For a more complicated story of open embeddings of C2 in certain affine
varieties see [16].
Example 4.3. Suppose k = k and let F be as in part a) of the proof of
4.2. Then V+ (F ) ' P1 ⇐⇒ d = 1 or d = 2. So not all embeddings of P1 in
P2 are equivalent (automorphisms of P2 preserve degree).
2. The case of affine spaces

“Affine algebraic geometry” deals with questions about affine spaces and
closely related algebraic varieties. The most important embedding problem
in this field is the following question.
Question 4.3. Is any closed embedding of Y = Am in X = An equivalent
to a linear embedding? We then call the embedding rectifiable.
We can put this in two ways: Let Y1 closed in X be given, Y1 ' Am .
(i) We fix coordinates x1, ..., xn on X beforehand and ask whether an
automorphism ϕ of X = Spec k[x1, ..., xn] can be found so that
ϕ(Y1 ) = V (ϕ∗(xm+1 ), ..., ϕ∗(xn )).
(ii) We ask whether coordinates x1, ..., xn on X can be found so that
Y1 = V (xm+1 , ..., xn).
4.4. (i) The AMS-theorem: This is true for m = 1, n = 2 (lines in the
plane) if char(k) = 0.
AMS refers to Abhyankar and Moh [2] and Suzuki [72], who indepen-
dently gave the first, quite different, proofs. There were many subsequent
other proofs based on again different ideas, e.g. [56, 59], the latter using
knot theory.
3. THE HIGH CODIMENSION CASE 115
(ii) This is not true even for m = 1, n = 2 , if char(k) > 0 (see Example
4.1 part (ii)). The classification of lines in the plane in this case is one of the
truly challenging remaining problems in 2-dimensional geometry.
(iii) There are no counterexamples known if char(k) = 0 and k = k.
On the other hand it has been pointed out in [68] that there are closed
embeddings R ⊆ R3 that represent knots and so cannot be rectified over R.
(iv) The high codimension theorem: This is true for n > 2m + 1, in
particular m = 1, n = 4.
Several, in principle quite similar, proofs [19, 36, 45, 71] were given, the
first two with slightly weaker bounds.
(v) Here are the next two (very challenging) open cases in zero charac-
teristic.
(v.1) m = 1, n = 3 (lines in space). See [11, 18, 19, 68] for some partial
results.
(v.2) m = 2, n = 3 (planes in space). See [46, 62, 65, 66] for some partial
results.
4.5. One can, of course, study the embedding problem in other categories,
for instance in complex analytic geometry. In that case there exist non rec-
tifiable lines in the plane, i.e. proper holomorphic embeddings of C into C2
that are not equivalent to a linear embedding. See [22], for instance.
4.6. An important generalization [54]of the AMS-theorem is the
Lin-Zaidenberg theorem: Let C be a topologically contractible curve
in X = C2. Then there are coordinates x, y for X such that C = V (xp − y q )
for some p, q ∈ N, GCD(p, q) = 1.
3. The high codimension case

This case is very different in spirit from, say, the case m = 1, n = 2, or
more generally the hypersurface case m = n − 1. The proofs establish good
properties of certain “generic” linear projections and are applicable to quite
general Y . The best result for non-singular Y is due to Kaliman [45] and
Nori (see [71]).
Theorem 4.7. Let Y be a non-singular affine variety of dimension m.
Then,
(i) Y admits a closed embedding in A2m+1 .
(ii) If Yi ⊆ X = An is closed, Yi ' Y, i = 1, 2 and n > 2m + 1, then
Y1 ∼ Y2 by a tame automorphism of An .
Proof. This is a sketch of the proof of Craighero [19] for (ii), the first
for a result of this nature, for the case Y = Am and n > 2m + 1 replaced by
n > 3m. Note that this includes the case m = 1, n = 4.
Step I. We look at Y 1 ⊆ Pn = X. The variety Z of lines in Pn meeting
Y 1 in at least 2 points or tangent to Y 1 has dimension ≤ 2m + 1. Hence
Z ∞ = Z ∩ X∞ (X∞ = hyperplane at infinity of X) has dimension ≤ 2m and
dimZ ∞ + m − 1 ≤ 3m − 1 ≤ n − 2 = dimX∞ − 1. Hence for a general linear
subspace L of X∞ of dimension m − 1 we have
(∗) L ∩ Z ∞ = ∅.
Let π be the projection out of such an L. (∗) implies that
(∗∗) any of the m-dimensional fibers of π (by definition they intersect X∞ in
L) meets Y1 in at most one point and is not tangent to Y1 .
After a linear change of coordinates we may assume that
L = V+ (X0 , X1 , ..., Xn−m ),
with V+ (X0 ) = X∞ . Then
π|X : X −→ X
is given by
(x1, ..., xn) 7−→ (x1, ..., xn−m, 0, ...0).
Let π 0 be the restriction of π to Y1 . By (∗∗), π 0 is injective and so is the
induced map on the tangent space at any p ∈ Y1 . Since Y 1 ∩ L = ∅, π 0 is a
proper map. It follows that π 0 is a closed embedding.
Step II. We need
Lemma 4.8. Let σ : k[x1, ..., xn] → B be a surjective homomorphism
(corresponding to a closed embedding of Spec B into An ) and put σ(xi ) = bi .
Suppose b1, ..., bn−m generate B over k, where m > 0. Let vj ∈ B, j = 1, ..., m.
Define τ : k[x1, ..., xn] −→ B by
τ (xi ) = bi , i = 1, ..., n − m
and
τ (xn−m+j ) = vj , j = 1, ..., m.
Then σ and τ are equivalent by an elementary automorphism.
Proof. Since B = k[b1, ..., bn−m], we can find polynomials Gi , Hi ∈
k[x1, ..., xn−m], i = 1, ..., m such that Gi (b1, ..., bn−m) = bn−m+i and Hi (b1 , ...,
bn−m ) = vi . Then
η : k[x1, ..., xn] −→ k[x1, ..., xn]
η(xi ) = xi, i = 1, ...., n − m
4. GENERAL AND GENERIC FIBERS 117
η(xn−m+j ) = xn−m+j + Gj (xi , ..., xn−m+j ) − Hj (xi , ..., xn−m+j ), j = 1, ..., m

gives the desired elementary automorphism.
To finish the proof of Theorem 4.7, we choose variables u1, ..., um for Y =
A and put B = Γ(Am ) = k[u1, ..., um]. A closed embedding Y ' Y1 ⊆ An
m
corresponds to choosing b01 , ..., b0n ∈ B so that B = k[b01, ..., b0n]. By Step I
there is a linear automorphism λ of A = k[x1, ..., xn] such that b1 , ..., bn−m
generate B, where bi = λ(b0i ). By the lemma there are elementary auto-
morphisms η1 , η2 transforming (b1, ..., bn) into (b1, ..., bn−m, u1 , ..., um) and
(b1, ..., bn−m , u1, ..., um) into (0, ..., 0, u1, ..., um) respectively. The composite
of λ, η1 and η2. gives a tame automorphism transforming the given embed-
ding into a standard linear one.
Exercise 97. Deduce the following from Lemma 4.8: Let
ρ : Am −→ An
be a closed embedding. Then the embedding
Am −→ An × Am
y 7−→ (y, 0)
is rectifiable. Work this out explicitly for Example 4.1 (ii). Note that we are
claiming that (y 4 + x + x6, z)k[x, y, z] is generated by a subset of a set of
variables for k[x, y, z], note that y 4 + x + x6 is a variable.
4. General and generic fibers

For simplicity we suppose k = k in this section. Let
k⊆B⊆A
be finitely generated k-algebras, with A a domain. We have the corresponding
morphism
f : X = Spec A −→ Y = Spec B
of affine k-varieties. The fiber fy of f over a closed point y ∈ Y is Spec A/my A,
where my ⊆ B is the maximal ideal corresponding to y. The generic fiber
of f , or the fiber over the generic point of Y , is Spec S −1 A, where S is
the multiplicative set B \ {0}. Note that this is an algebraic variety over
K = S −1 B, the field of quotients of B. We can also write S −1 A = K ⊗B A.
The fibers fy are affine varieties over B/my = k (since k = k).
We say f generically (resp. generally) has property P if the generic fiber
has property P (resp. fy has property P for y ∈ U, U a dense open subset
of Y ).
For example
Definition 4.9. X is generically (resp. generally) affine m-space over Y
if S −1 A ' K [m] (resp. A/my A ' k [m] , y ∈ U , U dense open in Y ).
Theorem 4.10. (i) Suppose X is generically affine m-space over Y . Then
it is so generally.
(ii) The converse of (i) is true if k = C and m = 1 or m = 2.
Proof. (i) Since A is finitely generated over B, we can find s ∈ S, i.e.
[m]
0 6= s ∈ B, such that As = Bs . Then U = Spec Bs is dense in Y and
A/my A = As/my As = k [m] for y ∈ U .
(ii) Without restriction on m it is true, at least under a mild assumption
on k (it holds for k = C), that there is a finite algebraic extension of fields
L ⊇ K such that L ⊗K S −1 A ' L[m] . See Corollary 4.33. The next result
then finishes the proof.
Theorem 4.11. (Absence of non-trivial separable forms of the affine line
and the affine plane) Let K be a field. Let D be a finitely generated K-
algebra and suppose L ⊇ K is separably algebraic and m = 1 or m = 2. If
L ⊗K D ' L[m] , then D ' K [m] .
For m = 1 the proof of 4.11 is a nice exercise in Galois theory, requiring
an application of both the additive and multiplicative version of Hilbert’s
Theorem 90. For two quite different proofs in case m = 2 see [44] and [64].
The corresponding question for m ≥ 3 is very interesting and very open.
Exercise 98. Let char(k) = 2 and f = y 4 + x + x6 ∈ A = k[x, y] as
in 1.2 (ii). Take B = k[f ] and S = B \ {0}. Then S −1 B = k(f )[x, y] '
k(t)[x, y]/(f + t) = E, t transcendental over k[x, y]. Then E k(t)[1], but for
e
suitable e, L ⊗k(f ) E ' L[1], where L = k(t1/2 ) is purely inseparable over
k(t).
So Theorem 4.11 is not true without the separability assumption. In Ex-
ercise 98 the generic fiber of f is a purely inseparable form of A1 . It is a much
worked on conjecture that this is always the case for a line V (f ) ⊆ A2 .
We establish a variant of the AMS-Theorem valid in arbitrary character-

istic.
Theorem 4.12. Suppose k = k. Let A be a finitely generated factorial
k-domain with A∗ = k ∗. Suppose f ∈ A and V (f ) is a generic line in X =
Spec A. Then A ' k [2] and V (f ) is a coordinate line in X, i.e. with S =
4. GENERAL AND GENERIC FIBERS 119
k(f ) \ {0} we have

S −1 A = k(f )[1] ⇒ A = k[f ][1].
In particular A ' k [2].
Proof. (a) Since A is factorial, we have k(f )∗ ∩ A = (S −1 A)∗ ∩ A =
A = k∗.
∗
(b) Let p be an irreducible element of k[f ]. If p = ab with a, b ∈ A, then

a, b are units in S −1 A = k(f )[1] and by (a), a, b ∈ A ∩ k(f ) = k[f ]. So p is
irreducible, hence prime, in A. It follows that pA ∩ k[f ] = pk[f ].
Write A = k[x1, ..., xn] and S −1 A = k(f )[t]. We may assume t ∈ A. We can
find pi ∈ S so that
pi xi ∈ k[f ][t].
Let p be an irreducible factor of p1 , say. Write
X
p1 x1 = aj tj , aj ∈ k[f ].
If p|aj in A, then by (b), p|aj in k[f ]. If this is the case for all j, we can
cancel
Q p from the above relation and reduce the number of prime factors in
pi . Otherwise in A/pA, X j
0= aj t
is a non-trivial relation for t over k = k[f ]/p. Hence t ∈ k, i.e.
t − a = pt0 with a ∈ k, t0 ∈ A.
P 0 0j
Clearly S −1 A = k(f )[t0]. Now p1 x1 = f (pt0 + a) = aj t , with aj ∈ pA for
j > 0. But then a0 ∈ pA as wellQ and we can again cancel p. By induction on
the number of prime factors in pi we see that we can assume
xi ∈ k[f ][t]
to begin with. Then A = k[f ][t].
Remark 4.13. Suppose char(k) = 0. In view of 4.12 we can interpret the
AMS-Theorem as asserting that a line in X = A2 is contained in a fiber of
an A1 -fibration of X, i.e. a morphism with general fiber isomorphic to A1.
In this form the theorem has been generalized in [33] to certain other affine
surfaces X.
The fiber dimension m is 1 in Theorem 4.12. It is clear from the proof that
the task there is to find a generic variable t in A that remains transcendental
over k in each fiber. The situation is considerably more complicated for
m ≥ 2. The task would be to find a generic set of variables t = (t1 , ..., tm)
in A that remain algebraically independent everywhere. The difficulty lies in
carrying out the last step of the proof, the modification of t to a suitable t0,
if this is not the case. See [67] for the case m = 2.
Exercise 99. Consider the hypersurface X = V (x + x2 y + z 2 + t3 ) ⊆ C4
and the morphism
f : X −→ C
given by x. The generic fiber of f is Spec k(x)[z, t] ' A2k(x) , whereas the
fiber f0 is the singular surface C × C, C = V (z 2 + t3) ⊆ C2 . Note that
f0 is homeomorphic to C2. X is what is known as an exotic affine space, an
algebraic variety diffeomorphic but not isomorphic to Cm . In particular Γ(X)
is factorial and has trivial units. See [47] and [49] for more on this.
An important positive result is the following from [67].
Theorem 4.14. Suppose char(k) = 0. Let B be a discrete valuation ring
with field of quotients K and maximal ideal m. Let A be a finitely generated
[2]
B-algebra such that A/mA ' (B/mB) and K ⊗B A ' K [2]. Then A ' B [2].
This result is easily extended to the case when B is a PID, but it is not
clear whether, or in what form, it extends to B of dimension ≥ 2. For a
specific open question see [75], in particular Property 5 and Property 6.
The following result from [46] for m = 2 is not quite as strong as Theorem
4.12 is for m = 1, but the best one can expect in the light of examples as in
Exercise 99. It establishes an important property of A3 that distinguishes it
from general exotic affine spaces.
Theorem 4.15. Let k = C and f ∈ A = C[3]. If f generally defines a
plane, then f generically defines a plane, i.e. k(f ) ⊗k[f ] A ' k(f )[2] , and f is
a variable in A.
In the proof it is shown that after a choice of generic variables special
fibers not isomorphic to C2 are similar to the one appearing in Exercise 99
and that the choice of variables can be improved by a suitable modification
process. Once no such fibers appear, all fibers are C2 and Theorem 4.14 com-
bined with 4.10(ii) gives the result.
The assumption that char(k) = 0 can’t be avoided in 4.15:

Exercise 100. Suppose char(k) = 2. Consider the hypersurface
X = V (xy + z 4 + t + t6 ) ⊆ A4
5. MORE ON N=3 121
(see 1.2(ii)) and the morphism

f : X −→ A1
given by x. The generic fiber of f is Spec k(x)[z, t] ' A2k(x), whereas the fiber
f0 is the surface C × A1 , A1 ' C = V (z 4 + t + t6 ) ⊆ A2 = Spec k[z, t].
This is a non-rectifiable embedding of A1 in A2 and it can be shown [3]
that A = Γ(X) 6= k[x][2]. It is not known whether X ' A3 . However [3],
X × A1 ' A4 . This in fact easily follows from Exercise 97. If X ' A3,
τ · (x, y, t, z) = (τ x, τ −1 y, t, z) defines a k ∗ -action on X. If X = A3 this action
is non-linearizable. (The fixed point set V (z 4 + t + t6) is not defined by a
variable in the quotient Spec k[z, t].) So we either have such an example, or a
counter example to the cancellation problem. It would be highly interesting
to settle which possibility occurs.
Remark 4.16. Polynomials f ∈ k[x, y] such that the generic fiber of
f : A2 −→ A1
is a rational curve over k(f ) (this means k(x, y) = k(f, g) for some g ∈
k(x, y)) have been extensively studied. See [16, 61], for example. They go
under the name of field generators or generically rational polynomials.
5. More on n=3
Let us begin with a very elementary and quite useful statement.
Exercise 101. Let S be a commutative ring and
X
α ∈ S[T ] ' S [1], α = αi T i , αi ∈ S.
Then
(i) α is nilpotent ⇐⇒ all αi are nilpotent.
(ii) α is a unit ⇐⇒ α0 is a unit and αi is nilpotent for i ≥ 1.
(iii) S[α] = S[T ] ⇐⇒ α1 is a unit and αi is nilpotent for i ≥ 2.
For the proof of the following proposition see [63]. It is a prototype for
many similar results.
P
Proposition 4.17. Let R be a commutative ring, b ∈ R, α = αi v i ∈
R[v] ' R[1] , αi ∈ R. Let H = bw + α ∈ R[v, w] ' R[2] , α1 a unit mod bR and
αi , i ≥ 2, nilpotent mod bR. Then there exists ϕ(T ) ∈ R[T ] such that
ϕ(α) ≡ v mod bR[v].
For any such ϕ,
ϕ(H) = v + bG with G ∈ R[v, w].
Moreover
R[G, H] = R[v, w].
A linear plane in A3 is an A2 defined by an equation that is linear in
a variable for A3. In [66] (for char(k) = 0) and [63] it is shown that such
planes are rectifiable. Here is a precise statement from [63]. The last part of
the proof is a direct consequence of 4.17.
Theorem 4.18. Let a, b ∈ A ' k [2], b 6= 0 and H = bw − a ∈ A[w] ' k [3].
If
(*) A[w]/HA[w] ' k [2],
then
(**) there exist u, v ∈ A such that A = k[u, v], b ∈ k[u] and a − v is
nilpotent mod bA.
Moreover, there exists G ∈ A[w] = k[u, v, w] such that
k[u, v, w] = k[u, G, H].
Conversely, if (∗∗) holds, then so does (∗).
This result
P hasi been considerably generalized in [65] to polynomials of the
form H = aiw such that the ai , i ≥ 1 have a non trivial common factor
b ∈ A. In another direction it has been generalized in [76] to polynomials
linear in variables w1, ..., ws and coefficients in k[x, y].
As mentioned before, the case of space lines (n = 3, m = 4) is not at all
settled. I do not even know of a counterexample to rectifiability in positive
characteristic.
Let
f : A1 = Spec k[t] −→ A3 = Spec k[x, y, z]
be given by
f (t) = (x(t), y(t), z(t)).
1
We put L = f (A ).
It had been proposed that f with the property that none of degt x, degt y,
degt z is in the semigroup generated by the other two are not rectifiable:
there is then no obvious elementary automorphism that will reduce one of
the degrees. We discuss an example [18] showing that this is not the case.
Note that f is a closed embedding precisely if
Rf = {F ∈ k[x, y, z]|F (x(t), y(t), z(t)) = t} =
6 ∅.
Moreover, Rf then is a coset of
I(L) = {H ∈ k[x, y, z]|H(x(t), y(t), z(t)) = 0}.
6. CURVES WITH ONE PLACE AT INFINITY 123
We note that if F ∈ Rf , then L meets V (F ) transversally in one point. It

is clear also that if L meets a plane P defined by a linear equation transver-
sally in one point, then f is rectifiable: If, say, P = V (x), then x(t) is linear
in t.
It follows that if L meets a rectifiable plane P transversally in one point,
then f is rectifiable. Conversely, if f is rectifiable, then there is an automor-
phism
(x, y, z) −→ (P (x, y, z), Q(x, y, z), R(x, y, z))
3
of A such that P (x(t), y(t), z(t)) = t. Here V (P ) is a rectifiable plane. We
therefore have the following.
Lemma 4.19. Let F ∈ Rf . Then f is rectifiable ⇐⇒ there exists H ∈ I(L)
such that P = F + H defines a rectifiable plane.
Consider now f given by
x(t) = t5 + t, y(t) = t4, z(t) = t3.
We have t = x(t) − x(t)y(t) + z(t)3, so we can start with F = x − xy + z 3.
V (F ) is not a plane, so we try to modify F by H ∈ I(L). This takes some
experimentation, but here is the result. We have x(t)y(t) + z(t)3 = t5 +
2t9 , x(t)2z(t) = t5 + 2t9 + t13, yz 3 = t13. So H = xy + z 3 − x2z − yz 3 ∈ I(L).
Now P = F + H = x + 2z 3 − x2 z − yz 3 is linear in y and satisfies (∗∗)
of Theorem 4.18 (with A = k[z, x]) and defines a rectifiable plane. So f is
rectifiable. As an exercise, construct an automorphism of A3 that rectifies f .
It will look quite complicated.
6. Curves with one place at infinity

Curves that miss just one non-singular point to be complete play an
important role in the study of affine curves. Lines and, over C, topologically
contractible curves, are important examples. The formal definition for plane
curves is as follows.
Definition 4.20. Let k be a field and F irreducible in k[ξ, η]. Put
A = k[ξ, η]/F = k[ξ, η]
and K = qt(A), the field of quotients of A.
We say F , or A, has one place at infinity, or more briefly is a one place curve,
if
(i) there is precisely one valuation subring D = DF of K/k such that D + A,
and
(ii) DF is residually rational over k.
Remark 4.21. Since K has transcendence degree 1 over k, a valuation

subring D 6= K of K/k is a discrete valuation ring (DVR). Let m be its
maximal ideal. Then D is residually rational if D/m = k. Note that m is
principal, and if m = τ D, then for 0 6= a ∈ K,
a = τ vD (a)u
with vD (a) ∈ Z, and u ∈ D∗ .
vD is the valuation associated to D. We put vF = vDF . The reader may
want to consult a standard textbook, e.g. [52] or [79], for facts on discrete
valuations.
Example 4.4. Suppose V (F ) is a line, i.e A = k[t] ' k [1]. Then F has
one place at infinity with DF = k[1/t](1/t) and vF = −degt .
Remark 4.22. Let C = V (F ) ⊆ A2 , C the closure of C in P2 and C̃ the

normalization of C in K. (C̃ is the complete regular curve determined by
F .) Then F has one place at infinity precisely when there is a unique q̃ ∈ C̃
lying above C \ C, and moreover q̃ is rational over k. In particular there is a
unique q ∈ C \ C, and q is rational over k. Note that this condition, which
we express as F has one point at infinity, is much weaker than having one
place at infinity.
The following is a key property of one-place curves.
Lemma 4.23. Suppose F has one place at infinity and let ζ ∈ A, ζ ∈
/ k.
0
Then A is integral over A = k[ζ] and
vF (ζ) = −[K : k(ζ)].
Proof. Let D be a discrete valuation subring of K/k and suppose D ⊇
A . If D = DF , then every discrete valuation subring of K/k contains A0.
0
Since their intersection is k (here we use that DF is residually rational), we

find ζ ∈ k and have reached a contradiction. Hence D 6= DF and D ⊇ A.
This says that A is integral over A0. Also, DF is the unique discrete valuation
subring of K/k lying over k[1/ζ](1/ζ) and hence −VF (ζ) = VF (1/ζ) = [K :
k(ζ)].
4.24. Let F ∈ k[ξ, η] have one place at infinity.
(i) By 4.23, either ξ ∈ k and F = aξ + b with a, b ∈ k and a 6= 0 or η is
integral over k[ξ] and hence
F = a0η n + a1 (ξ)η n−1 + ... + an (ξ)
6. CURVES WITH ONE PLACE AT INFINITY 125
with 0 6= a0 ∈ k and
degη (F ) = n = [k(ξ, η) : k(ξ)] = −vF (ξ).
We will assume a0 = 1.
(ii) By Definition 4.20, F has one point at infinity, which we will assume is
on the line {η = 0}. This means that
the degree form of F is η n .
Then
deg(F ) = degη (F ) = n > degξ (F ) = m.
So either m = 0 and F = aη + b with a, b ∈ k and a 6= 0 or m > 0 and F
is monic in ξ, i.e. the degree coefficient in ξ is a non-zero constant. (We may
assume it is 1 if we like.)
4.25. We keep the notation of 4.24.
(i) We introduce local coordinates at p∞ = (0, 1, 0), the point at infinity
of F (see the appendix to this chapter),
X = ξ −1 , Y = ηξ −1 .
Then for G ∈ k[ξ, η],
G∞ = Gξ −deg(G)
is the local equation at p∞ of V (G). We have
G∞ ∈ k[X, Y ] ⊆ k[[X]][Y ] ⊆ k[[X, Y ]].
(ii) In view of Lemma 4,23, f = F∞ is monic in Y and of the form
f = Y n + a1(X)Y n−1 + ... + an (X)
with X|ai (X) for i = 1, ..., n.
An f ∈ k[[X]][Y ] is as in 4.25(ii) is called a monic distinguished polyno-
mial, or a polynomial in Weierstrass form.
4.26. If p∞ = (0, 1, 0) is a point at infinity of G ∈ k[ξ, η], then one can
show that the places at infinity of G at p∞ , or, in the notation of Remark
4.22, the p̃ ∈ C̃ lying above p∞ , are in one to one correspondence with the
irreducible factors of G∞ in k[[X]][Y ], or equivalently in k[[X, Y ]]. This gives
the following basic result:
Proposition 4.27. With notation as in 4.24, F ∈ k[ξ, η] has one place at
infinity if and only if (see 4.35 for details) f = F∞ is irreducible in k[[X, Y ]]
and residually rational over k.
4.28. Let the situation be as in 4.24. Let G ∈ k[ξ, η] and suppose ζ =

G ∈ A is not constant. Then for p ∈ V (F, G) ⊆ A2 the local intersection
multiplicity i(V (F ), V (G))p is defined as the k-dimension of the local ring of
p on A2 divided by local equations for F and G at p. Let C̃ be as in 4.22.
Then ζ defines a divisor (ζ) on C̃. For p ∈ V (F, G) let p̃ = {p1 , ..., ps} be
the points of C̃ lying above p. It is a key property of the local intersection
multiplicities (it in fact characterizes them) that
i(V (F ), V (G))p = degree of the part of (ζ) supported by p̃.
Hence
i(VP
(F ), V (G))A2
= p∈V (F,G) i(V (F ), V (G))p
= deg((ζ)0)
= [K : k(ζ)]
= −vF (G).
Here (ζ)0 is the divisor of zeros of (ζ). We write i(F∞ , G∞ ) for the local
intersection multiplicity of V (F ) and V (G) at p∞ .
By Bezout’s theorem,
i(V (F ), V (G))A2 + i(F∞ , G∞ ) = deg(F )deg(G).
So
i(F∞, G∞ ) = VF (G) + ndeg(G).
In particular
i(F∞ , Y ) = vF (η) + n = n − m and i(F∞, X) = n.
Let F have one place at infinity. We will study translates of F , generic

or otherwise, by using the powerful Theorem 4.46. Set
G = F + t, f = F∞, g = G∞ .
Here we can take t ∈ k or in any extension field of k. In particular, t could
be transcendental over k. It is clear that F and G have the same point at
infinity and that f and g are Weierstrass polynomials of the same degree. So
W (f ) = W (g). If char(k) = 0, one can use the theory of approximate roots
of Abhyankar and Moh [1, 2] to prove that the assumption i(f, g) > λ(f ) is
satisfied as well. See [62]. We obtain the following basic result.
Theorem 4.29. Suppose char(k) = 0 and let F have one place at infinity.
Let G = F + t be a translate of F , where t could be transcendental. Then G
has one place at infinity. Moreover, g = G∞ passes through each infinitely
near multiple point of f = F∞ with the same multiplicity as f .
APPENDIX CHAPTER 4 PART 1: NOTATION 127
Let the notation be as in Remark 4.22 and set n = deg(F ). Suppose

C = V (F ) is non-singular.
P Then by the genus formula the genus of C̃ is
1
2
((n − 1)(n − 2) − µ i (µ i − 1)), where µ1 , µ2 , ... is the multiplicity sequence
of F at infinity. See 4.45. The point here is that the genus depends on n and
the µi > 1 only. By the last part of Theorem 4.29, the complete non-singular
curves C̃F and C̃G determined by F and G have the same genus. So in case
V (F ) is a line, C̃G is a curve of genus 0 and the point at infinity is a rational
point. It follows that V (G) is the complement of a rational point in P1k(t) and
hence is a line. So we have
Corollary 4.30. Suppose char(k) = 0 and let V (F ) be a line. Then
V (F ) is a generic line.
Together with Theorem 4.12 this gives a proof of the AMS-theorem.
Remark 4.31. Theorem 4.29 fails in positive characteristic. Ganong has
shown in [31] that F + t, t transcendental over k, has ”geometrically” one
place at infinity in the sense that there is one DVR at infinity, but with
residue field a purely inseparable and possibly non-trivial extension of k.
Moreover, F + c can have several places at infinity for c ∈ k.
Appendix chapter 4 part 1: Notation

k will denote a field, k its algebraic closure.
If A is a k-algebra, X = Spec A is the corresponding affine k-scheme.
We then also write A = Γ(X) = ring of regular functions on X. If E ⊆
A is a subset, V (E) ⊆ X denotes the zero-locus of E. Formally, V (E) =
Spec A/I, I= ideal generated by E.
A∗ will denote the group of units of A.
k [n] will denote the polynomial algebra of dimension n over k and Ank =
Spec k [n] will denote affine n-space over k . We suppress k if it is clear which
field is meant.
Choosing variables for k [n] means to pick precisely n elements x1 , ..., xn in
k so that k [n] = k[x1, ..., xn]. We will then also say that we have introduced
[n]
coordinates x1, ..., xn on X = An . If f ∈ A = k [n] , we call f a variable if

A = k[f ][n−1] .
If k = k we can identify An with k n = {(a1, ..., an)|ai ∈ k}.
We say Y = Spec B is an algebraic variety if B is finitely generated (or
of finite type or affine) over k, say by b1, ..., bn, and reduced. We then have a
surjective homomorphism
β : A = k[x1, ..., xn] −→ B
β(xi) = bi
which defines a closed embedding of Y into X = An . (Consider this as the
definition of ”closed embedding”.)
We recall that a B-algebra A is said to be finite over B if A is a finite
B-module.
Pn denotes projective n-space. We write its homogeneous coordinate ring
as k[X0, X1 , ..., Xn ]. Note that in contrast to the affine situation, the homo-
geneous coordinates Xi are determined up to an invertible linear transfor-
mation. We have k(Pn ) = field of rational functions on Pn = {P/Q|P, Q ∈
k[X0 , X1 , ..., Xn] homogeneous of the same degree, Q 6= 0}.
If E ⊆ k[X0 , X1 , ..., Xn] is a set of homogeneous elements, V+ (E) ⊆ Pn
denotes the zero-locus of E.
V+ (X0 ) ' Pn−1 and Pn \ V+ (X0 ) ' X = An = Spec k[x1, ..., xn] with
xi = Xi /X0 . We call X∞ = V+ (X0 ) the hyperplane at infinity of X.
To study subvarieties Y of X = An , we often consider Y , the closure of
Y in X = Pn , and Y∞ = Y ∩ X∞ , the locus at infinity of Y . There is a
1 − 1-correspondence between closed irreducible subvarieties of X = An and
closed irreducible subvarieties of X = Pn not contained in X∞ , which I will
describe for hypersurfaces. Let
f = F0 + ... + Fd ∈ k[x1, ..., xn], Fi homogeneous of degree i, Fd 6= 0.
Put
f+ = X0d f (X1 /X0 , ..., Xd/X0 )
= X0d F0 + X0d−1 F1 (X1 , ..., Xn) + .... + Fd (X1 , ..., Xn ).
Note f+ ∈ k[X0 , X1 , ..., Xn] is homogeneous and that X0 - f+ . We have
Y = V+ (f+ ) and Y∞ = V+ (X0 , Fd).
For example, if n = 2, Y = V (f ) is an affine plane curve, Y a projective
plane curve and Y∞ consists of finitely many points, the points at infinity of
Y . In the other direction, let F (X0, ..., Xn ) ∈ k[X0, ..., Xn ] be homogeneous
of degree d with X0 - F and put
fa = F (1, x1, ..., xn) ∈ k[x1, ..., xn].
We can write
F = X0d F0 + X0d−1 F1 + .... + Fd
with Fi ∈ k[X1, ..., Xn ] homogeneous of degree i, Fd 6= 0. Then
fa = F0 + F1(x1, ..., xn) + ... + Fd (x1, ..., xn).
APPENDIX CHAPTER 4 PART 2: MORE ON FIBERS 129
If Z = V+ (F ), then Z ∩ X = V (fa ) and Z ∩ X∞ = V+ (X0 , Fd ).

Pn is covered by open sets U0 , ..., Un with
X0 Xi−1 Xi+1 Xn
Ui = Spec k[ , ..., , , ..., ] ' A2 .
Xi Xi Xi Xi
Let F ∈ k[x1, ..., xn]. To study Y = V (F ) at infinity, i.e. to study Y , we
calculate
fi (y1, . . . , yn ) = f+ (y1, . . . , yi−1, 1, yi , . . . , yn ), i = 0, . . . , n
T
and note that Y Ui = V (fi ). Of course f0 = f .
In case n = 2, we find that P2 is covered by the affine planes
U0 = Spec k[x, y], U1 = Spec k[1/x, y/x], U2 = Spec k[x/y, 1/y]
with origins respectively at (1, 0, 0), (0, 1, 0), (0, 0, 1). If f ∈ k[x, y] has degree
d, then
f1 (u, v) = u−d F (u, v), f2(u, v) = u−d F (u, v).
Appendix chapter 4 part 2: More on general and generic fibers

Let k be a field and
k⊆B⊆A
k-algebras. For s ∈ Spec B we write
As = A ⊗B κ(s)
where κ(s) = qt(B/js), the field of quotients of B/js , where js is the prime
ideal corresponding to s. So Spec As is the scheme-theoretic fiber of
Z = Spec A −→ Spec B = S
over the generic point of the closed subset {s} ⊆ S. The following comparison
result for general and generic fibers was worked out by H. Kraft and P.
Russell.
Theorem 4.32. Suppose
(∗) k is algebraically closed and of infinite transcendence degree over the
prime field.
Let B be a finitely generated k-domain and A ⊇ B, A0 ⊇ B finitely generated
B-algebras. Suppose there is a non-empty Zariski open U ⊆ Spec B such that
for all closed points s ∈ U there exists a κ(s)-isomorphism
ϕs : A0s −→ As .
Then there exist 0 6= f ∈ B and a finite Bf -algebra C, C ⊇ Bf , and an

isomorphism of C-algebras
ϕ : A0f ⊗Bf C −→ Af ⊗Bf C.
Proof. Since we are free to localize B, A and A0 at 0 6= g ∈ B such that
Spec Bg ⊆ U , we may assume Spec B = U .
We can find a subfield k0 ⊆ k finitely generated over the prime field and
k0 -algebras B0 , A0, A00 of finite type such that
B = B0 ⊗k0 k, A = A0 ⊗k0 k, A0 = A00 ⊗k0 k.
Let K0 = qt(B0). By the assumptions (∗) on k there exists a k0 -embedding
K0 ,→ k.
The k0 -homomorphism
B0 ,→ K0 ,→ k
induces a k-homomorphism
η : B = B0 ⊗k0 k −→ k
and gives a closed point s ∈ Spec B with js = Ker(η). Note that js ∩ B0 =
js ∩ K0 = 0.
Now by assumption
A0 ⊗B k = A0s ' As = A ⊗B k.
We have
A ⊗B k = (A0 ⊗B0 B) ⊗B k = A0 ⊗B0 k = (A0 ⊗B0 K0 ) ⊗K0 k.
Similarly
A0 ⊗B k = (A00 ⊗B0 K0 ) ⊗K0 k.
We can now find a K0 -algebra D ⊆ k of finite type such that
(A0 ⊗B0 K0 ) ⊗K0 D ' (A0 ⊗B0 K0 ) ⊗K0 D.
Here we can replace D by D/m = L, where m is a maximal ideal of D. Hence
A0 ⊗B0 L = A00 ⊗B0 L,
where L is a field of finite degree over K0 . We can now replace L by a B0 -
subalgebra C 0 ⊆ L of finite type. Then by the Noether normalization theorem
(see 3.48) there exists 0 6= f ∈ B0 such that C0 = Cf0 is finite over B0f . Now
C = C0 ⊗k0 k meets the requirements of the theorem.
APPENDIX CHAPTER 4 PART 3: PLANE ALGEBROID CURVES 131
Corollary 4.33. Under the assumptions of 4.32 the generic fibers of

Spec A −→ Spec B and Spec A0 −→ Spec B
become isomorphic over a finite extension of qt(B).
Appendix chapter 4 part 3: Plane algebroid curves

Let k be a field and consider g ∈ k[[X, Y ]]. Many facts on power series are
used and mentioned below without proof and a textbook, e.g. [79], should
be consulted.
4.34. (i) We write
g = G0 + G1 + ... with Gν homogeneous of degree ν.
Let
µ = min{ν|Gν 6= 0}.
We call µ the multiplicity and Gµ the leading form of g.
If µ > 0, Spec k[[X, Y ]]/(g), or just g, is often called a plane algebroid curve.
Of course g is a unit if and only if µ = 0.
(ii) For g, h ∈ k[[X, Y ]], the intersection multiplicity is defined to be
i(g, h) = dimk k[[X, Y ]]/(g, h).
It is finite precisely when f and h have no non-unit common factor.
4.35. Suppose now f ∈ k[[X, Y ]] is irreducible. Put R = k[[X, Y ]]/(f )
and L = qt(R), the field of quotients of R. Let R̃ be the integral closure of
R in L.
(i) Then R̃ is a discrete valuation ring (DVR) with associated valuation v,
say, and maximal ideal
m = {x ∈ R̃|v > 0}.
We have R̃ = {x ∈ L|v ≥ 0}.
(ii)We say f is residually rational over k if R̃/m = k. Then R̃ is a complete
discrete valuation ring with residue field k and hence a power series ring in
one variable over k, which we can write as
R̃ = k[[τ ]],
where τ is any element of value 1 in L. Moreover, v is the same as τ − order.
(iii) For g ∈ k[[X, Y ]],
i(f, g) = v(g) (g = g mod f ).
Elements of k[[X, Y ]] of multiplicity 1 are analytic variables. In fact
Lemma 4.36. Let f = F1 + ..., g = G1 + ... be elements of k[[X, Y ]] with

F1, G1 linearly independent. Then k[[X, Y ]] = k[[f, g]].
Lemma 4.37. Let f = Fµ + ..., g = G1 + ... be elements of k[[X, Y ]],
with µ(f ) = µ > 0, µ(g) = 1. Then i(f, g) = µ if G1 - Fµ and i(f, g) > µ if
G1 |Fµ .
Proof. By Lemma 4.36 we can assume g = X. Then k[[X, Y ]]/(f, g) '
k[[Y ]]/f with f = f (0, Y ) = Y ν u, where u is a unit and ν = µ if X - Fµ and
ν > µ otherwise.
For f as in 4.37, the linear factors of Fµ are called the tangents of f .
(One could extend this definition to non linear irreducible factors). For a
polynomial p = Pµ + ... + Pd ∈ k[X, Y ], Pi homogeneous of degree i, vanishing
at the origin one defines the tangents at the origin in a similar fashion. Again
they are characterized by having local intersection multiplicity with p at the
origin larger than the minimum for a linear polynomial. An irreducible p can
have many tangents. This is quite different for power series.
Lemma 4.38. Let f ∈ k[[X, Y ]] be irreducible and residually rational.
Then the leading form Fµ is a power of a linear form.
Proof.
(a) Suppose Fµ = Gν Hλ with Gν , Hλ relatively prime. Let Ln denote the
vector space of forms of degree n. By unique factorization and an elementary
dimension count, the map
Ln−ν × Ln−λ −→ Ln
sending (H, G) to Gν H + Hλ G is surjective. It is then straightforward by
induction on l to construct gl = Gν + ... + Gν+l , hl = Hλ + ... + Hλ+l such
that gl hl agrees with f up to terms of order µ + l. Then
f = (Gν + Gν+1 + ...)(Hλ + Hλ+1 + ...).
(b) Suppose Fµ = Qs , where Q is irreducible of degree d > 1. By Lemma
4.37, i(f, X) = µ = i(f, Y ). So if x, y are the images of X, Y in R, then
in the notation of 4.35, v(x) = v(y) and v(y/x) = 0. Hence y/x ∈ R̃. Set
R0 = R[y/x] ⊆ R̃ and let P be the kernel of the map k[[X, Y ]][T ] −→ R0 that
sends T to y/x. Clearly f ∈ Q and XT − Y ∈ Q. Moreover, f ≡ X µ g modP
with g = Q(1, T )s + Xg 0 , g 0 ∈ k[[X, Y ]][T ]. Since X ∈
/ P, g ∈ P . It follows
that (X, Q, P ) = (X, Y, Q) ⊇ P is a maximal ideal. If m0 is its image in R0 ,
we have
R0 /m0 ' k[T ]/Q(1, T ) % k
and f is not residually rational over k.
Remark 4.39. Let p ∈ k[X, Y ] be as above and irreducible and set

A = k[X, Y ]/p. Then the irreducible factors of p in k[[X, Y ]] are in one-one
correspondence with the DV R0 s of the field of quotients of A centered at
(0, 0). Note that by 4.38 there is at least one such factor for each irreducible
factor of Pµ .
Let f ∈ k[[X, Y ]] be irreducible and residually rational. We will study f
through the process of quadratic transformations, or blow-ups, along f . See
[62] for details. Let
ϕf : k[[X, Y ]] −→ R ⊆ R̃ = k[[τ ]]
be the quotient map described in 4.35. We set x = ϕ(X), y = ϕ(Y ). We
assume x 6= 0.
Definition 4.40. Put p = i(f, Y ) = v(y), c = i(f, X) = v(x). We distin-
guish three cases.
(1) If p > c, then y/x ∈ R̃ and v(y/x) > 0. We set (x∗ , y ∗) = (x, y/x)
and (X ∗ , Y ∗ ) = (X, Y/X).
(2) If p < c, then x/y ∈ R̃ and v(x/y) > 0. We set (x∗, y ∗ ) = (y, x/y)
and (X ∗ , Y ∗ ) = (Y, X/Y ).
(3) If p = c, then y/x ∈ R̃, v(y/x) = 0 and there is a unique α ∈ k such
that v(y/x − α)) > 0. We set (x∗, y ∗) = (x, y/x − α) (X ∗ , Y ∗) =
(X, Y/X − α).
4.41. We have k[[X, Y ]] ⊆ k[[X ∗, Y ∗ ]] and if g ∈ k[[X, Y ]], then
g = X ∗µ(g) g̃ with g̃ ∈ k[[X ∗, Y ∗ ]].
Moreover f˜ is irreducible in k[[X ∗, Y ∗ ]] and generates the kernel of the map
ϕf˜ : k[[X ∗, Y ∗ ]] −→ R0 = k[[x∗, y ∗]] ⊆ R̃
that sends X ∗ to x∗ , Y ∗ to y ∗ .
4.42 (Noether’s formula). We have µ(f ) = min{p, c} = v(x∗ ) and if

g ∈ k[[X, Y ]], then
˜ g̃).
i(f, g) = v(ϕf (g)) = µ(g)v(ϕf (X ∗ )) + v(ϕf˜(g̃)) = µ(f )µ(g) + i(f,
We call g̃ the strict transform of g at the first infinitely near point of f
and call µ(g̃) the multiplicity of g at this point. We can continue this process
inductively, defining the successive infinitely near points of f , and the strict
transforms and multiplicities of f and g at these points. We can also see this
as inductively constructing rings
k[[x, y]] = k[[x1, y1]] ⊆ k[[x2, y2 ]] ⊆ ... ⊆ R̃ = k[[τ ]], (xi+1, yi+1 ) = (x∗i , yi∗).
We will see below that R̃ = k[[xi, yi ]] for i >> 0.
The strategy now is to handle questions about f , in particular about
intersection multiplicities, inductively. An example is the following direct
consequence of Noether’s formula
X
i(f, g) = µi (f )µi (g),
the sum extended over all infinitely near points of f . (Unless f |g, g will pass
through only finitely many of these, i.e. µi (g) = 0 from some point on, and the
sum will be finite.) We note that the sequence of the µj (f ) is non-increasing,
and so will eventually stabilize at some i with value δ, say. If we are in case
(1) or (2) of 4.40 and c - p or p - c, further quadratic transformations will
eventually reduce the multiplicity. We have therefore constructed (xi , yi ) =
(u, w) so that, say, δ = v(u) and δ|v(yj ) for j ≥ i. This says that for s > 0, w
is of the form ws us plus a polynomial in u, with v(ws ) > 0. Hence w ∈ k[[u]].
By 4.40, x and y are polynomials in xi , yi and hence x, y ∈ k[[u]]. If δ > 1,
we have k((x, y)) = k((u)) $ k((τ )) = L = k((x, y)). Hence δ = 1 and u = τ
up to a unit. We record this as:
4.43. By the process of successive quadratic transformations along f we
construct an element τ of value 1 in R̃.
P
4.44. Let g = aj (X)Y j ∈ k[[X, Y ]] and suppose X - g. Then the
Weierstrass degree of g is defined to be
w = W (g) = min{j ; X - aj (X)}.
Then by the Weierstrass preparation theorem there is a unit u ∈ k[[X, Y ]]
such that ug is a monic distinguished polynomial in Y of degree w, i.e.
ug = Y w + Xb1 (X)Y n−1 + ... + Xbw (X) with bj (X) ∈ k[[X]].
4.45. Let f ∈ k[[X, Y ]] be irreducible and residually rational. Let
µ1 ≥ µ2 ≥ ...
be the sequence of multiplicities of f at its successive infinitely near points.
We put
X
µ = min{µj |µj > 1} and λ = λ(f ) = µ + µ2j .
{j|µj >1}
The following is an important irreducibility and rationality criterion for

power series. See [62].
Theorem 4.46. Let f, g ∈ k[[X, Y ]] with X - f, X - g and suppose that
f is irreducible and residually rational and that
W (g) = W (f ) and i(f, g) > λ.
Then g is irreducible and residually rational. Moreover, g passes through
all infinitely near multiple points of f , in fact
if µi > 1, then µi (g) = µi .
One can consider λ as a kind of local self intersection number for f . To
understand it better one needs the theory of characteristic approximations
to f . These become a particularly powerful tool when combined with the
theory of approximate roots of Abhyankar and Moh [1, 2].
CHAPTER 5
Bifurcation set of the global Milnor fibration
Ha Huy Vui
1. Introduction and definitions

The purpose of this chapter is to report on some results on bifurcation
set of the global Milnor fibration defined by a polynomial mapping. These
results are closely related to the Jacobian conjecture.
Let us first look at some examples:
Example 5.1. Let f : C2 → C, (x, y) 7→ xy. The singular fiber F0 =
f −1 (0) is homeomorphic to C ∪ C.
Let t 6= 0, Ft = f −1 (t) and Φ : C∗ → Ft , ϕ(x) = (x, xt ). Φ is a complex
analytic isomorphism. In particular, Φ is a homeomorphism. Therefore
H1 (Ft, C) = H1 (C∗, C) = H1 (S1, C) = C
and
H1 (F0, C) = H1∗ (C ∪ C, C) = 0.
Example 5.2. Let f : C2 → C, (x, y) 7→ x2 y −x. f has no singular points.
F0 ≈ C ∪ C∗ (disjoint union). If t 6= 0, Ft ≈ C∗ (using the parametrization
technique as in Example 5.1 ). We see that
H0 (F0, C) = C2 , H0 (Ft, C) = C
but
H1 (F0, C) = H1 (Ft , C) = C.
In this case, the first homology group could not distinguish the special fiber
F0 from the general fiber Ft .
To understand what happens in the second example, we have to compare
the fibers F0 and Ft in a neighborhood of infinity. Namely, if Dr is a disc
with r sufficiently large, then F0 ∩ Dr and Ft ∩ Dr are homeomorphic, but
F0 \ Dr and Ft \ Dr are not.
Now let us to recall a general result of R.Thom:
137
138 5. BIFURCATION SET OF THE GLOBAL MILNOR FIBRATION
Theorem (R.Thom). Let f : Cn → Ck be a polynomial mapping. There

is an algebraic subset B in Ck , such that the mapping
f : Cn \ f −1 (B) −→ Ck \ B
defines a C ∞ -locally trivial fibration.
Let us denote by Bf the smallest of all the sets B of Thom’s theorem.
Then it is called the bifurcation set of f and the corresponding fibration
f : Cn \f −1 (Bf ) −→ Ck \Bf the global Milnor fibration.
The bifurcation set can be decomposed into two types of critical val-
ues (critical value is the “analytic” version of “singularity”): those that are
critical values, and those that are “critical at infinity”. Let us first give a
definition:
Definition 5.1. Let f : Cn → Ck be a polynomial mapping.
(i) A point y0 ∈ Ck is said to be a regular value at infinity of f , if there exists
a compact set K in Cn and a number δ > 0 such that the mapping
f : f −1 (Dδ ) \ K → Dδ
defines a trivial fibration, where Dδ = {y ∈ Ck | ky − y0 k < δ}.
(ii) If y0 is not a regular value at infinity of f , then it is said to be a
critical value, corresponding to the singularities at infinity of f .
In the examples 5.1 and 5.2, B(f ) = {0}. The value 0 is a critical value
for f = xy (as f −1 (0) is not smooth), but it is not so for f = x2y − x .
Put
Bf := K0 (f ) ∩ B∞ (f ),
where K0 (f ) is the set of critical values and B∞ (f ) is the set of critical values
corresponding to the singularities at infinity of f .
Problem. How to characterize the points of the set B∞ (f )?
This is a challenging problem of singularity theory. Until now it is solved
only for a few cases.
2. Case n = 2
In the example f (x, y) = x2y − x, we have seen that the group H1 cannot
distinguish the special fiber from the general fiber, but the group H0 can do
this. In fact, we have the following simple characterization of critical values,
corresponding to the singularities at infinity of polynomials in two complex
variables.
2. CASE n = 2 139
Theorem 5.2. [72], [34] Let f : C2 → C be a polynomial in two complex

variables. Let t0 be a regular value of f . Then the value t0 belongs to the set
B∞ (f ) if and only if the Euler characteristic χ(f −1 (t0)) of the fibre f −1 (t0)
is different from the Euler characteristic of the general fiber f −1 (t).
Let us firstly to recall how to calculate the Euler characteristic of (non
singular) affine curve.
Let C = {f (x, y) = 0} be an affine plane curve. Without loss of generality,
we can assume that the polynomial f (x, y) is of the form
f (x, y) = xd + a1(y)xd−1 + · · · + ad (y),
where the degree of f is d. In this case, the restriction π| C of π : C2 →
C, (x, y) → y, on C is proper. Let {(x1, y1), (x2 , y2), . . . , (xs , ys )} be the set
of critical points of π|C . Take a point y0 ∈ / {y1, y2 . . . , ys } in the plane C
and joint y0 with yi by a path Ti such that Ti has no self-intersection point
and Ti ∩ Tj = {y0}. The deformation retract of C into T1 ∪ . . . ∪ Ts can be
lifted by π| C into a deformation retract of C in πC−1(T1 ∪ ... ∪ Ts ). We can see
πC−1 (T1 ∪ . . . ∪ Ts ) is homotopically equivalent to a “graph”, whose vertices
are d points in πC−1 (y0 ) and s points (x1 , y1), ..., (xs, ys ).
There are ri edges which intersect at the vertex (xi, yi ), where ri is the
multiplicity of xi as a root of f (x, yi) = 0. We have
χ(C) = χ(graph) = d + s − (r1 + · · · + rs ).
Proof of Theorem 5.2. Assume that χ(f −1 (t0)) = χ(f −1 (t)). From
the above method, it follows that there are r > 0, δ > 0 such that the system
of equations
f (x, y) = t,
∂f
(x, y) = 0,
∂x
has no roots in f −1 (Dδ ) \ {y 6 r}.
Then, it is easy to construct a diffeomorphism trivializing the fibration
f : f −1 (Dδ ) \ {y 6 r} −→ Dδ .
Thus, the value t0 is regular at infinity of f.
Let f (x, y) = xd + a1 (y)xd−1 + · · · + ad (y). Consider f − t as an element of
C[y, t][x]. Denote by ∆(y, t) = disc(f (x, y)−t), the discriminant of f (x, y)−t.
We can write
∆(y, t) = a0(t)y s + a1 (t)y s−1 + · · · .
We have
Corollary 5.3. For a polynomial f : C2 → C, we have B∞ (f ) = {t ∈

C|a0(t) = 0}.
In turns out that for a polynomials in two complex variables one can
characterize critical values corresponding to singularities at infinity in terms
of the rate of gradient functions in neighborhood of infinity. This rate is
measured by the Lojasiewicz numbers at infinity.
Let f ∈ C[x1, . . . , xn ], t0 ∈ C, Dδ = {t||t − t0| < δ}. Put
ϕδ (r) = inf (k∇f (x)k)
kxk=r,x∈f −1 (Dδ )
where ∇f (x) is the gradient of f at x.

Definition 5.4. The number
ln ϕδ (r)
L∞,t0 (f ) := lim lim
δ→0 r→∞ ln r
is called the Lojasiewicz number at infinity of the fiber f −1 (t0).
Remark 5.5. (i) The inequality L∞ (t0) > 0 means that the following
Fedoryuk’s condition holds:
(F) For any sequence (xk1 , . . . , xkn ) → ∞ with the property that f (xk ) →
t0, the sequence k∇f (xk )k does not tend to 0.
(ii) The inequality L∞,t0 (f ) > −1 is equivalent to the fact that the fol-
lowing Malgrange’s condition holds:
(M) For any sequence {xk } ⊆ Cn , kxk k → ∞ with the property that
f (xk ) → t0, the sequence k∇f (xk )k · kxk k does not tend to 0.
Theorem 5.6. [35] Let f (x, y) be a polynomial over C. The following
conditions are equivalent
(i) t0 ∈
/ B∞ (f );
(ii) L∞,t0 (f ) < −1;
(iii) L∞,t0 (f ) < 0.
Proof. (Due to [51]). We may assume that
f (x, y) = a0xd + · · · , d := deg f.
We say that a fraction series
a(y) = c1y q1 /k + c2 y q2/k + · · · , q1 > q2 > · · · , k ∈ N
∂f
is a Puiseux root at infinity of ∂x
= 0 if the series a(tk ) converges at
(x, y)
infinity and ∂f
∂x
(a(y), y) = 0.
We shall show that each of conditions (i)-(iii) is equivalent to the following
condition
2. CASE n = 2 141
∂f
(iv) There is a Puiseux root at infinity of ∂x
(x, y) = 0 such that
f (a(y), y)) − t0 → 0, as y → ∞.
(i) ⇐ (iv) follows from the proof of Theorem 5.2.

(iv) ⇐ (ii): If ∂f
∂x
(a(y), y) = 0 and
f (a(y), y) − t0 = cy α + terms of degrees less than α, α < 0, c 6= 0,
then
∇f (a(y), y) = (0, αcy α−1 + · · · ).
Hence |y|k∇f (a(y), y)k → 0 as y → ∞.
Since f (x, y) is monic in x, there exist c1 , c2 > 0 such that
c1 k(a(y), y)k ≤ |y| ≤ c2k(a(y), y)k, |y| 1.
Thus we have (ii).
(ii) ⇐ (iii): Clearly.
(iii) ⇐ (iv): Assume that the Fedoryuk condition (F) is not satisfied at
t0. Using the curve selection lemma, there exists an analytic curve
λ : x = c1sn1 + . . . , y = s−N ,
where s → 0, N > 0, n1 < n2 < . . . (ni need not be positive). We must have
n1 + N ≥ 0, since xn /yn is bounded. We can rewrite λ as a fractional power
series
λ : x = c1 y −n1 /N + c2 y −n2 /N + . . . , −N ≤ n1 < n2 . . . .
Let’s apply the change of variables
X = x − λ(y), Y = y −1 .
Put X
M(X, Y ) = f (X + λ(1/Y ), 1/Y ) − t0 = cij xi y j/N .
For each cij 6= 0, let us plot a dot at (i, j/N ), called a Newton dot. The set
of Newton dots is called the Newton diagram of M. Clearly, it has at most
finitely many dots lying on or below the X-axis. Moreover, there is one dot
at (d, 0).
The assumption f (λ(y), y) − t0 → 0 means that M(0, Y ) → 0 as Y → 0.
All Newton dots of M(0, Y ) lie above the X-axis.
If there is no dot of M(x, y) lying on X = 1, then
∂M ∂f
(0, Y ) = (λ(x), y) = 0
∂X ∂x
and we have (iv). If it is not the case, let (1, h1 ) denote the lowest Newton
dot on X = 1. We must have h1 > 0 since otherwise ∂M ∂X
(0, Y ) = ∂f
∂x
(λ(x), y)
does not tent to zero when y tends to infinity.
Now the idea is to use the Newton dots on or bellow the X-axis to “swal-
low” (1, h1 ).
Let’s consider this idea on the example
M(X, Y ) = Y 3 − 2XY + X 3 Y −1 + X 4 .
The dot (1, 1) represent −2XY . We use the dot (3, −1) (which represents
X 3 Y −1 ) to swallow (1, 1).
p dH(z)
Choose a non-zero root c = 2/3 of = 0 with H(z) = z 3 − 2z
dz
and γ = (2/3)Y . We see that the lowest Newton dot on X = 1 of
p
Mf(X, Y ) = M(X + 2/3 Y, Y )
is higher than (1, 1). P
In the general case, let M(X, Y ) = cij X i Y j/N and let EH 0 be the
∂M
highest Newton edge of . The dot (0, h1 ) is lying on EH 0 . Let us collect
∂X
all the terms whose derivatives lie on EH 0
X
ϕH (X, Y ) := cij X i Y j/N , (i − 1, i/N ) ∈ EH 0 .
d
Let c be a non-zero root of ϕH 0 (z, 1) = 0. In the expansion of ϕH 0 (X +
dz
cY tan θH 0 , Y ), where θH 0 is the angle between EH 0 and the X−axis, the term
d
XY h1 has coefficient 0, since ϕH 0 (c) = 0. Denote by γ1 (Y ) := cY tan θH , we
dz
∂M
say that γ1 (Y ) is the result of the sliding of λ along .
∂X
We see that the lowest Newton dot on X = 1 of M1(X, Y ) = M(X +
γ1 (Y ), Y ) is higher than (1, h1 ).
On X = 0, all dots remain above the X−axis. A recursive sliding λ → γ1 →
γ2 → . . . , will then yields a series γ, for which
f
M(X, Y ) = M(X + γ(Y ), Y )
has no dots on X = 1, and dots on X = 0 all lie above the X−axis.
Thus
∂ M̃ ∂M
(0, Y ) = (γ(Y ), Y ) = 0.
∂X ∂X
The series γ̃(y) := γ(1/y) satisfies the condition (iv).

2. CASE n = 2 143
It is important to know when the set B∞ (f ) is empty. For this purpose,

we use the Lojasiewicz number at infinity of f . Let
ϕ(r) = inf k∇f (x)k.
kxk=r
Definition 5.7. The number

ln ϕ(r)
L∞ (f ) = lim
r→+∞ ln r
is called the Lojasiewicz number at infinity of f .

Note that L∞ (f ) is the largest number α such that the inequality
k∇f (x)k ≥ ckxkα
holds since c > 0 for every x lying outside a certain compact set in Cn .
Theorem 5.8. Let f ∈ C[x, y], the following two conditions are equivalent
(i) L∞ (f ) < −1.
(ii) B∞ (f ) 6= ∅.
Proof. (i) =⇒ (ii). Put

V = {(x, y)k∇f (x, y)k = min k∇f (u, v)k}.
k(u,v)k=k(x,y)k
The set V is semi-algebraic, obviously unbounded. Let ϕ : [0, ] −→ Cn be a

meromorphic curve, such that ϕ(τ ) ∈ V, kϕ(τ )k → ∞.
We see that

d
≤ < ∇f (ϕ(τ )), dτ >
dϕ(τ )
dτ f (ϕ(τ ))
dϕ(τ )
≤ k∇f (ϕ(τ ))kk k ∼ k∇f (ϕ(τ ))k · kϕ(τ )k · τ −1 .
dτ
Here A ∼ B means that A/B lies between two positive constants. Hence
d

τ f (ϕ(τ )) ≤ k∇f (ϕ(τ ))kkϕ(τ )k
dτ
This implies that there is t0 ∈ C, such lim f (ϕ(τ )) = t0 and L∞ (t0 ) < −1.
τ →0
Hence t0 ∈ B∞ (f ).
The implication (i) =⇒ (ii) is clear
Let us indicate a relation between the set B∞ (f ) with the Jacobian Con-
jecture. Let F := (f, g) : C2 → C2 be a polynomial mapping with nonzero
Jacobian. In addition, assume that the set B∞ (f ) is empty. According to
Abhyankar-Moh’s theorem, we can choose coordinates in C2 such that f = x.

For this choice
1 0
J (f, g) = = gy = c.
gx gy
Thus g(x, y) = cy + k(x) and (x, y) 7→ (x, cy + k(x)) is a bijection.
Remark 5.9. (i) In [53] Lê Dung Tráng and C. Weber described very
effective way to compute the set B∞ (f ) of a polynomial f : C2 → C via
resolution of indeterminacy of f at infinity.
(ii) Some characterizations of the set B∞ (f ) in the case n = 2 can be
found in [25].
(iii) If f has isolated singularities at infinity, then A. Parusinski [57] has
shown that the three following conditions (a)-(c) are equivalent (a) t0 ∈
B∞ (f ), (b) χ(f −1 (t0)) − χ(f −1 (t)) 6= 0 and (c) L∞ (t0) < −1.
3. Rabier-Jelonek Fibration Theorem

Now let us to consider the general case f : Cn → Ck with k > 1.
In contrast to the case n = 2, k = 1, here we can only prove that the
set B∞ (f ) is contained in the set of points in Ck , at which the so-called
generalized Malgrange’s condition is not satisfied.
Rabier and Gaffney numbers. Denote the set of linear maps from X
to Y by L(X, Y ) and by Σ(X, Y ) ⊆ L(X, Y ) the set of non-surjective linear
maps.
Definition 5.10. Let A ∈ L(X, Y ). Set
ν(A) = inf kA∗(Φ)k,
kΦk=1
where A∗ ∈ L(Y ∗ , X ∗ ) is the adjoint operator and Φ ∈ Y ∗

The number ν(A) can be characterized as follows:
ν(A) = dist(A, Σ) = inf kA − Bk.

B∈Σ
We call ν(A) the Rabier number of A.

Let A ∈ L(k n , k m ) (where n ≥ m). Let (aij ) be the matrix of A. Let MI ,
where I = (i1, ..., im), denotes an (m × n) minor of (aij ) given by columns
indexed by I. Let MJ (j) denote an (m − 1) × (m − 1) minor given by columns
3. RABIER-JELONEK FIBRATION THEOREM 145
indexed by J and by deleting the j th row (if m = 1 we put MJ (j)=1). By

the Gaffney number of A we mean the number
|MI |
g(A) = max{min }.
I J⊆I MJ (j)
(Here we consider only numbers with MJ (j) 6= 0, if all numbers MJ (j) are
zero, we put g(A) = 0).
We can prove that
ν(A) ∼ g(A).
Definition 5.11. Let k = C or k = R. Let f : k n → k m be a polynomial
mapping. Put
K∞ (f ) := {y ∈ k m : ∃ xl → ∞ s.t. f (xl ) → y & |xl kν(df (xl )) → 0}.
Note that by virtue of the results above we can replace the function ν by
the function g.
Theorem 5.12. ([30], [42], [58]) Let K0 (f ) be the set of critical values
of f and B(f ) be the set of bifurcation value of f . Then
B(f ) ⊆ K(f ) = K0 (f ) ∪ K∞ (f ).
Proof. Let a 6∈ K(f ). Without loss of generality we can assume that a = 0.
We have that a 6∈ K0 (f ) and a 6∈ K∞ (f ) . This implies that there are
R > 0, > 0, η > 0 such that for every x with kxk ≥ R and kf (x)k < η, we
have
|MI |
max{min kxk } > .
I J⊆I |MJ (j)|
Moreover, there is ω > 0, such that for every x with kxk ≤ R and kf (x)k < η
we have maxI |MI (x)| ≥ ω. Let U = {y ∈ k m : kyk < η} and let Γ = f −1 (0).
We show that f −1 (U ) ∼
= Γ × U and f is a projection Γ × U 3 (γ, u) 7→ u ∈ V .
Indeed, let us define sets
UI = {x ∈ f −1 (U ) : minJ⊆I kxk |M|MJ (j)|
I|
= if kxk = R; |MI (x)| =
ω if kxk 5 R}
and
VI := {x ∈ f −1 (U ) : minJ⊆I kxk |M|MI (j)|
I|
5 /2 if kxk ≥ R; |MI (x)| ≤
ω/2 if kxk 5 R}.
The sets VI and UI are disjoint. Consequently, there is a C ∞ function δI :
k n → [0, 1] which is equal to 1 on UI and 0 on VI . It is easy to check
P that
the sets HI = {x : δI 5 0} cover the set f −1 (U ). Now take δ := I δI and
let ∆I = δI /δ.
Take y = (y1, .., ym) ∈ U . Take the index I = (1, .., m) and we consider
the following (formal) system of equation
X
n
∂f1
(x)Vj (x) = y1
j=1
∂xj
...
X
n
∂fm
(x)Vj (x) = ym
j=1
∂xj
Vm+1 = 0
Vn = 0
Using the Cramer’s rule we can solve this system, let Mj (i) = Mk , for J =
I\k, we have
X
m
V1 (x) = (−1)1+k yk M1k /MI
k=1
...
X
m
Vm (x) = (−1)m+k yk Mmk /MI
k=1
Vm+1 = 0
...
Vn = 0
We have df (VI (y, x)) = y. In an analogous way we can define VI for an

arbitrary index I = (ii1 , .., iim ). P
Now consider a vector field V (y, x) = I ∆I (y, x) in the domain f −1 (U ).
By the construction, we have
|V (x)| ≤ 2mn/kxk for kxk ≥ r.
Let us consider the differential equation

0
x (t) = V (y, x(t)),
x(0) = γ,
3. RABIER-JELONEK FIBRATION THEOREM 147
where γ ∈ Γ. Note that

X
df (V (y, x)) = df ( ∆I VI (y, x))
I
X
= ∆I df (VI (y, x))
I
X
= ( ∆I )y = y.
I
Consequently, if x(t, y, γ) is a solution of the system, then f (x(t), y, γ) =

yt. Since y ∈ U , we have that the trajectory x(t, y, γ), t ∈ [0, t0) does not
intersect the boundary ∂f −1 (v) for every 0 ≤ t0 ≤ 1 + δ for some δ > 0.
Now we prove that the trajectory x(t, y, γ) is defined on the whole of [0, 1].
In fact, we have
Assertion. Let U ⊆ k n be an open set and V : U → k n be a smooth
mapping. Assume that for kxk large enough, we have
kV (x)k < Mkxk.
Let γ ∈ U and let x(γ, t), t ∈ [0, t0] be a solution of the following system
0
x (t) = V (x(t)),
x(0) = γ.
Then this trajectory is bounded. In particular, this trajectory either is defined
for every t > 0 or it intersects the boundary ∂U of U .
Proof. By assumption we have
kV (x)k < Mkxk for kxk > R.
Let r(t) = kx(γ, t)k2. We will show that the function r is bounded. As-
sume on the contrary that there is a sequence of points ti → t0 such that
r(ti ) → ∞ as ti → t0.
Let N be any regular value of r with N > R. The pre-image r−1 (N ) is
an at most countable set w = w1 , w2, . . .. For every wi ∈ W two cases are
possible:
(a) r(t) ≤ N for every t ∈ [wi, wi+1 ].
(b) r(t) ≥ N for every t ∈ [wi, wi+1 ]
(if wi+1 is not defined then we put t0 = wi+1 ).
Of course only the case (b) is interesting. Hence, assume that (b) holds.
By the assumption for a number t ∈ [wi, wi+1 ] we have kV (x(t))k ≤ Mkx(t)k.
Consequently, for every such t, we have:

0
r (t) ≤ kx(t)kkV (x)k < 2Mkxk2 = 2Mr(t).
0
It means that ln(r(t)) < 2M and finally r(t) ≤ r(wi ) exp(2M(t − wi )) ≤
N exp(2M(t−wi)). Thus the set r(t), wi < t < t0 is bounded, a contradiction.
Let us return to the proof of the theorem. By assertion, the trajectory
x(t, y, γ) is defined on the whole of [0, 1]. Since f (x(t, y, γ)) = yt, we have
that the phase flow x(t, y, γ), t ∈ [0, 1], transforms f −1 (0) = Γ into f −1 (y).
Let
ϕ : Γ × U 3 (γ, y) → x(1, y, γ) ∈ f −1 (V ).
It is easy to see that ϕ is a diffeomorphism. Thus 0 6∈ B(f ).
Let us consider a polynomial mapping f : k n → k n . We say that f is
proper at a point y ∈ k n , if there exists an open neighborhood U of y such
that the restriction f|f −1 (U ) : f −1 (U ) → U is a proper map.
We denote by Jf the set of points at which f is not proper. Then we
always have the inclusion K∞ (f ) ⊆ J (f ).
Proposition 5.13. Let f : Cn → Cn be a dominant polynomial mapping.
Then K(f ) = B(f ).
Proof. We have B(f ) ⊆ K(f ) and by definition K(f ) ⊆ K0 ∪ Jf . On
/ B(f ) if and only if #f −1 (y) = µ(f ),
the other hand, it is easy to see that y ∈
where µ(f ) is a geometric degree of f . It is equivalent to
B(f ) = K0 (f ) ∪ J = K0 (f ) ∪ K∞ (f ) = K(f ).

4. Jelonek set
We call Jf the Jelonek set of f. It turns out that if f is a polynomial
mapping from Rn to Rm , or from Cn to Cn , then Jf has a very useful property.
Namely, Jf is R-uniruled (for real polynomial maps) and C-uniruled (for the
complex case). In fact, we have
Theorem 5.14. ([39], [41]) Let f : Rn −→ Rn be a polynomial non-
constant mapping. Then the set Jf is closed, semi-algebraic and for every
non-empty connected component S ⊆ Sf , we have 1 ≤ dim S ≤ n − 1.
Moreover, the set Sf is R-uniruled. It means that for every point a ∈ Sf ,
there is a non-constant polynomial mapping ϕ : R −→ Sf such that ϕ(0) = a.
A curve Γ is called an affine parametric line if it is affine and there is a
dominant polynomial mapping f : C −→ Γ .
4. JELONEK SET 149
Theorem 5.15. ([37]) Let f : Cn −→ Cn be a polynomial mapping. Then

the set Jf of points at which the mapping is not proper is either empty or it
is a hypersurface. Moreover, the variety Sf is C-uniruled. It means, that for
every point a ∈ Sf there is an affine parametric line Γa in Sf going through
a.
Let us consider a polynomial mapping f : Rn −→ Rn . In this case, the
codimension of Jf may equal n − k for every 0 < k < n.
Theorem 5.16. ([41]) Let f : Rn −→ Rn be a real polynomial map-
ping with nonvanishing Jacobian. If codimJf ≥ 3 then f is a bijection (and
consequently Jf = ∅).
Proof. Let us consider the sets X = Rn \Jf and Y = Rn \f −1 (Jf ). Since
the mapping f is a local homeomorphism, codim f −1 (Jf ) = codim Jf ≥ 3.
Since Jf and f −1 (Jf ) are semialgebraic sets, we can show that Y is connected
(for this it is enough that codim Jf ≥ 2) and X is simply connected (we
have to use the fact that codim Jf ≥ 3). Since the mapping f : Y → X
is proper and unramified, it gives a topological covering. In particular, it
is a homeomorphism as X is simply connected. Consequently, the general
fiber of the mapping f is one point. Since the mapping f is unramified, this
implies that f is an injection. So, f is a bijection by the well-known fact that
“Injectivity ⇒ Surjectivity”.
Remark 5.17. The example of Pinchuk shows that there are real poly-
nomial mappings with nonvanishing Jacobian and with codim Jf = 1. Hence
the only interesting case is that of codim Jf = 2.
Conjecture 5.18 (Jelonek conjecture). ([41]) Let f : Rn −→ Rn be a
real polynomial mapping with nonvanishing Jacobian. If codim Jf ≥ 2 then
f is a bijection (and consequently Jf = ∅).
Remark 5.19. By [41, Theorem 4.2], this conjecture is true in dimension
two. Consequently, the first interesting case is n = 3 and dim Jf = 1.
The Jelonek Conjecture is closely connected with the Jacobian Conjec-
ture.
Proposition 5.20. ([41]) The Jenolek Conjecture in dimension 2n im-
plies the Jacobian Conjecture in (complex) dimension n.
Proof. Let f : Cn −→ Cn be a polynomial mapping with a non-zero
Jacobian. We can treat the mapping f as a real polynomial mapping f :
Rn −→ Rn . Assume that f is not an isomorphism. In particular the mapping
f cannot be finite. Thus by Theorem 5.15, we get that the set Jf has a
complex codimension 1, hence it has a real codimension 2. Consequently, if
the Jelonek Conjecture holds in dimension 2n, this gives a contradiction.
The second application to the Jacobian Problem of results about the
Jelonek set is a solution by Jelonek of the problem posed by Arno van den
Essen and Shpilrain.
Let p ∈ C[x1, . . . , xn ] be a polynomial. We say that p is a coordinate
polynomial, if it can be included in a generating set from n elements of the
algebra C[x1, . . . , xn ].
If p is a coordinate polynomial, then it is irreducible and its zero-set is
isomorphic to Cn−1 .
Let ϕ : C[x1, . . . , xn ] −→ C[x1, . . . , xn ] be a polynomial automorphism,
then if f is a coordinate polynomial, then so is ϕ(f ). Conversely, Arno van
den Essen and Vladimir Shpilrain have stated the following
Problem. Is it true that every endomorphism of C[x1, . . . , xn ] taking
any coordinate polynomial to a coordinate one is actually an automorphism?
Let us recall some fact about C-uniruled varieties. For an affine variety
X ⊆ Cn , we say that X has s non-uniruled components at infinity, if the
variety cl(X)\X (where cl denotes the projective closure in Pn (C)) has s
non-uniruled irreducible components.
It is proved by Jelonek that:
• A C-uniruled affine irreducible variety X ⊆ Cn has at most one
non-uniruled component at infinity.
• Every smooth hypersurface S ⊆ Pk (C) of degree r > k is not unir-
uled. A projective variety X is said to be uniruled if it is of positive
dimension and for a generic point in X there exists a rational curve
in X through this point.
Solution of the Problem. We prove something more general.
Theorem 5.21. ([37]) Let Φ be an endomorphism of C[x1, . . . , xn ] and D
be a natural number. Assume that Φ takes any coordinate polynomial p with
deg p ≥ D, to a coordinate one. Then, Φ is an automorphism.
Sketch of the proof. Let Φ be an endomorphism of C[x1, . . . , xn ] and Φ
takes any coordinate polynomial p with deg p ≥ D, to a coordinate one. Φ
induces a polynomial mapping f : Cn −→ Cn , Φ(p) = p ◦ f . It is enough to
show that f is an automorphism of Cn .
We do it in two steps. First we show that f has an invertible Jacobian.
Finally, we show that f is a finite mapping.
4. JELONEK SET 151
Assertion. Let p1 , . . . , pn ∈ C[x1, . . . , xn ]. Assume that for every non-

zero vector (a1, . . . , an) ∈ Cn , the polynomial a1p1 +· · · +an pn is a coordinate
polynomial, then det(∂pi/∂xj ) ∈ C∗ .
∂pi
Proof. Let J (x) = det( (x)). Assume that J (x) = 0 for some x ∈ Cn .
∂xj
Then there is a non-zero vector (a1, . . . , an ) such that
a1∇p1 + · · · + an ∇pn = 0.
P
P that ∇( aipi (x)) = 0, which contradicts the fact that the poly-
This means
nomial aipi is a coordinate polynomial.
Let f = (f1 , . . . , fn ) be as before. Let h1 , . . . , hn be polynomials of degrees
≥ D, which give a polynomial system of coordinates.
Denote h := (h1 , . . . , hn ). For every non-zero vector (a1, . . . , an ) the poly-
nomial a1h1 + . . . an hn is a coordinate polynomial of degree ≥ D. By the
assumption, the polynomial a1h1 ◦ f + . . . an hn ◦ f is also a coordinate poly-
nomial. Thus we have J (h ◦ f ) = J (h) · J (f ) ∈ C∗ . Therefore J (f ) ∈ C∗ .
Now we shall show that f is a finite mapping.
The case n = 1 is trivial.
Suppose that n ≥ 2. Let us assume on the contrary that the mapping f
is not an automorphism, in particular, it cannot be finite. Let V be a non-
empty irreducible component of the set Jf . Hence, V is a hypersurface. Let us
assume for a moment that there is a coordinate polynomial π of degree ≥ D,
such that π −1(0) ∩ V is not a C-uniruled variety.
Let us consider the mapping
f 0 : f −1 (π −1 (0)) 3 x −→ f (x) ∈ π −1 (0).
Note that π −1 (0) and f −1 (π −1(0)) are isomorphic to Cn−1 . So, the mapping
f 0 can be treated as an (unramified) mapping Cn−1 → Cn−1 .
One can show that π −1 (0)∩V is contained in the set Jf 0 of points at which
the mapping f 0 is not proper. Since the latter variety is not C-uniruled, we
have a contradiction.
The fact that the polynomial π exists can be shown explicitly by using
the properties that we have mentioned before, namely: Every C-uniruled
affine irreducible variety X ⊆ Cn has at most one non-uniruled component
at infinity and every smooth hypersurface S in Pk (C) of degree r > k is not
uniruled.
Bibliography
[1] S.S. Abhyankar, Lectures on expansion techniques in algebraic geometry, Tata

Lecture Notes 57, Tata Inst. of Fundamental Research, Bombay, 1977.
[2] S.S. Abhyankar, T.T. Moh, Embeddings of the line in the plane, J. Reine Angew.
Math. 276 (1975), 148-166.
[3] Asanuma, T., Non-linearizable algebraic group action on An, J. Algebra 166, No.
1 (1994), 72-79.
[4] K. Adjamagbo, J.-Y. Charbonnel and A. van den Essen, On ring homomorphisms
of Azumaya algebras, preprint April 2005.
[5] K. Adjamagbo and A. van den Essen, A short proof of the equivalence of the
Dixmier, Jacobian and Poisson Conjectures, (to appear).
[6] M. Artin, Algebra. Prentice Hall Inc., 1991.
[7] M. F. Athiyah and I. G. Macdonald, Introduction to Commutative Algebra.
Addison-Wesley Publ. Comp., Reading Mass. 1969.
[8] H. Bass, The Jacobian Conjecture and Inverse degrees, Arithmetic and Geometry,
Prog. Mathematics No. 36,(1983), 65-75.
[9] H. Bass, E. Connell and D. Wright, The Jacobian Conjecture: Reduction of Degree
and Formal Expansion of the Inverse, Bull. Amer. Math. Soc. 7 (1982), 287-330.
[10] A. Belov and M. Kontsevich, Jacobian Conjecture is stably equivalent to Dixmier
Conjecture, preprint April 2005.
[11] S.M. Bhatwadekar, A. Roy, Some results on embeddings of a line in 3-space, J.
Algebra 142 (1991), no. 1, 101-109.
[12] J.-E. Björk, Rings of Differential Operators, North-Holland Library, Vol. 21, 1979.
[13] M. de Bondt and A. van den Essen, A reduction of the Jacobian Conjecture to
the symmetric case, Proc. Amer. Math. Soc. 133(2005), no.8, 2201-2205.
[14] M. de Bondt and A. van den Essen, Singular Hessians, J. Algebra 282 (2004),
no.1, 195-204.
[15] M. de Bondt and A. van den Essen, Nilpotent symmetric Jacobian Matrices and
the Jacobian Conjecture,II, J. Pure Appl. Algebra 196 (2005), no.2-3,134-148.
[16] P. Cassou-Nogues and P. Russell, Birational endomorphisms of the affine plane
and affine-ruled surfaces, in Affine Algebraic Geometry, Osaka University Press,
2007, 57-106.
[17] E. H. Connell, A K-theory for the category of projective algebras, J. Pure Appl.
Algebra 5 (1974), 281-292.
[18] P.C. Craighero, About Abhyankar’s conjectures on space lines, Rend. Sem. Mat.
Univ. Padova 74 (1985), 115-122.
153
154 BIBLIOGRAPHY
[19] P.C. Craighero, A result on m-flats in An k , Rend. Sem. Mat. Univ. Padova 75
(1986), 39-46.
[20] S. Cynk and K. Rusek, Injective endomorphisms of algebraic and analytic sets,
Ann. Polon. Math. 56 (1991), No.1, 29-35.
[21] H. Derksen, Inverse degrees and the Jacobian Conjecture, Comm. Algebra, 22,
(1994),No.12, 4793-4794.
[22] H. Derksen and F. Kutzschebauch, Nonlinearizable holomorphic group actions,
Math. Ann. 311 (1998), 41-53.
[23] J. Dixmier, Sur les algèbres de Weyl, Bull. Soc. Math. France, 96 (1968), 209-242.
[24] W. Dicks, Automorphisms of the polynomial ring in two variables, Publ. Sec.
Mat. Univ. Autònoma Barcelona 27 (1983), no. 1, 155-162.
[25] A. Durfee, Five definitions of critical point at infinity, Progress in Mathematics,
162 (1998), Birkhäuser, Basel, 345-360.
[26] D. Eisenbud, Commutative Algebra with a View Towards Algebraic Geometry.
Graduate Texts in Math. vol. 150, Springer Verlag 1995.
[27] A. van den Essen, Polynomial automorphisms and the Jacobian conjecture,
Progress in Mathematics, vol. 190, Birkhäuser Verlag, Basel, 2000.
[28] A. van den Essen, The sixtieth anniversary of the Jacobian Conjecture: a new
approach, Ann. Polon. Math. 76, No.1-2 (2001), 77-87.
[29] A. van den Essen, On Bass’ inverse degree approach to the Jacobian Conjecture
and exponential automorphisms, Contemp. Math 264 (2000), 207-214.
[30] T. Gaffney, Fibers of polynomial mappings at infinity and a generalized Malgrange
condition, Compositio Math. 119 (1999), no. 2, 157-167.
[31] R. Ganong, On plane curves with one place at infinity, J. Reine Angew. Math.
307/308 (1979), 173-193.
[32] C. Godbillon, Géometrie différentielle et Mécanique Analytique, Hermann Paris,
1969.
[33] R. Gurjar, K. Masuda, M. Miyanishi and P. Russell, Affine lines on affine surfaces
and the Makar-Limanov in variant, to appear in Can. J. Math.
[34] Hà Huy Vui and D. T. Lê, Sur la topologie des polynôme complexes, Acta Math.
Vietnam. 9 (1984), 21-32.
[35] Hà Huy Vui, Nombres de Lojasiewicz et singularités à l’infini des polynômes de
deux variables complexes, C.R. Acad. Sci. Paris, t.311, Serie I (1990), 429-432.
[36] Z. Jelonek, The extension of regular and rational embeddings, Math. Ann. 277
(1987), 113-120.
[37] Z. Jelonek, Testing sets for properness of polynomial mapping, Math.Ann.315
(1991), 1-35 .
[38] Z. Jelonek, The Jacobian conjecture and and the extensions of polynomial em-
beddings, Math. Ann. 294 (1992), 289-293.
[39] Z. Jelonek, The set of points at which a polynomial map is not proper, Ann.Pobn.
Math. LVIII.3 (1993), 259-266 .
[40] Z. Jelonek, A solution of the problem of van den Essen and Shpilrain, J. Pure
Appl. Algebra, 137 (1999), 49-55 .
[41] Z. Jelonek, A geometry of real polynomial mappings, Math. Z. 239 (2002), 321-
333 .
BIBLIOGRAPHY 155
[42] Z. Jelonek, On the generalize critical values of a polynomial mapping,

Manuscripta Math. 110 (2003), 145-157 .
[43] W. Heinrich and E. Jung, Über ganze birationale Transformationen der Ebene,
J. Reine Angew. Math. 184 (1942), 161–174.
[44] T. Kambayashi, On the absence of non-trivial forms of the affine plane, J. Alge-
bra 35 (1975), 449-456.
[45] S. Kaliman, Extensions of isomorphisms between affine algebraic subvarieties of
kn to automorphisms of kn, Proc. Amer. Math. Soc. 113 No.2 (1991), 325-333.
[46] S. Kaliman, Polynomials with general C2 fibers are variables, Pacific J. Math.
203 (2002), 161-189.
[47] S. Kaliman and L. Makar-Limanov, On the Russell-Koras contractible threefolds,
J. Algebraic Geom. 6 (1997), 247-268.
[48] O. Keller, Ganze Cremona-Transformationen, Monathsh. Math. Phys., 47
(1939), 299-306.
[49] M. Koras and P. Russell, Contractible threefolds and C∗ -actions on C3, J. Alge-
braic Geom. 6 (1997), 671-695.
[50] W. van der Kulk, On polynomial rings in two variables, Nieuw Arch. Wisk. (3)
1 (1953), 33–41.
[51] T-C. Kuo and A. Parusinski, Newton polygon relative to an arc, Real and Com-
plex singularities, 76-93, Chapman & HallCRC Res. Notes Math, 412, (2000).
[52] S. Lang, Algebra, Graduate Texts in Mathematics, Springer.
[53] Lê Dung Tráng and C. Weber, Polynômes à fibres rationnelles et conjecture
jacobienne à 2 variables, C. R. Acad. Sci. Paris Sr. I Math. 320 (1995), no. 5,
581-584.
[54] V.Y. Lin and M. Zaidenberg, An irreducible simply connected curve in C2 is
equivalent to a quasihomogeneous curve, Soviet Math. Dokl. 28 (1983), 200-204.
[55] H. Matsumura, Commutative algebra, Cambridge Univ. Press, Cambridge, 1989.
[56] M. Miyanishi, Analytic irreducibility of certain curves on a non-singular affine
rational surface, Proc. Kyoto Conf. on Algebraic Geometry, 1977.
[57] A. Parusinski, On the bifurcation set of complex polynomial with isolated singu-
larities at infinity, Compositio Math. 97 (1995), No. 3.
[58] P. J. Rabier, Ehresmann’s Fibration and Palais-Smale conditions for morphisms
of Finsler manifolds, Ann. of Math. 146 (1997), 647-691 .
[59] L. Rudolph, Embeddings of the line in the plane, J. Reine Angew. Math. 276
(1982), 113-118.
[60] K. Rusek and T. Winiarski, Polynomial automorphisms of Cn, Univ. Iagel. Acta
Math. 24 (1984), 143-149.
[61] P. Russell, Field generators in two variables, J. Math. Kyoto Univ. 15 (1975),
555-571.
[62] P. Russell, Hamburger-Nöther expansions and approximate roots of polynomials,
Manuscripta Math. 31 (1980), 25-95.
[63] P. Russell, Simple birational extensions of two dimensional affine rational do-
mains, Compositio Math. 33 (1976), 197-208.
156 BIBLIOGRAPHY
[64] P. Russell, Some formal aspects of the theorems of Mumford-Ramanujam, Proc.

Int. Coll. on Algebra, Arithmetic and Geometry 2000, Tata Inst. of Fundamental
Research, Narosa Publishing 2002, 557-584.
[65] P. Russell and A. Sathaye, On finding and cancelling variables in k[X, Y, Z], J.
Algebra 57 (1979), 151-166.
[66] A. Sathaye, On linear planes, Proc. Amer. Math. Soc. 56 (1976), 1-7.
[67] A. Sathaye, Polynomial rings in two variables over a DVR: a criterion, Invent.
Math. 74 (1983), 159-168.
[68] A. R. Shastri, Polynomial representations of knots, Tohoku Math. J. (2) 44
(1992) no.2, 11-17.
[69] Ivan P. Shestakov and Ualbai U. Umirbaev, The Nagata automorphism is wild,
Proc. Natl. Acad. Sci. USA 100 (2003), no. 22, 12561-12563 (electronic).
[70] S. Smale, Mathematical Problems for the next century, Math. Intelligencer, 20
(1998), No.2, 7-15.
[71] V. Srinivas, On the embedding dimension of an affine variety, Math. Ann. 289
(1991), 125-132.
[72] M. Suzuki, Propriétés topologiques des polynômes de deux variables complexes,
et automorphismes algébriques de l’espace C2 , J. Math. Soc. Japan 26 (1974),
241-257.
[73] R. Thom, Ensembles et Morphismes stratifie, Bull. Amer. Math. Soc. 75 (1969),
249-312.
[74] Y. Tsuchimoto, Endomorphisms of Weyl algebra and p-curvatures, Osaka J. Math
42 (2005), 435-452.
[75] S. Venereau, in: Open problems in affine algebraic geometry, Contemp. Math.
369 (2005), 17-23.
[76] S. Venereau, Hyperplanes of the form f1 (x, y)z1 + .... + fk (x, y)zk + g(x, y) are
variables, Can. Math. Bull. 48 no.4 (2005), 622-635.
[77] S. Wang, A Jacobian criterion for separability, J. Algebra 65 (1980), 453-494.
[78] A. Yagzhev, On Keller’s Problem, Siberian Math. J.21 (1980), 747-754.
[79] O. Zariski, P. Samuel, Commutative Algebra, Vol. II, Graduate Texts in Mathe-
matics, Springer.
[80] W. Zhao, Hessian nilpotent formal power series and their deformed inversion
pairs, preprint 2004, to appear.
List of notations
B∞ (f), 138 U B(n), 27

Bf , 138 BAn()R, 8
F ∗ , 21 GAn(R), 3
F∞ , 125 MAn (R), 3
G∞ , 125 supp F , 10
Hes(f), 31 dϕx , 95
Jf , 148 divD, 29
K0 (f), 138 e∗i ∧ e∗i+n , 40
R∗ , 1 expD, 29
Tx X, 82 f+ , 128
U B(n), 27 g(A), 145
V+ (E), 128 k(Pn ), 128
X∞ , 128
Xnorm , 104
Xsing , 85
Afn(R), 4
AlgC (O(Y ), O(X)), 62
Ared , 77
AnR, 1
∆(X), 68
Derx (O(X)), 82
EAn (R), 7
GA0n(R), 5
GAon(R), 4
Mor(X, Y ), 62
Pn , 128
Tn (R), 8
Trn(R), 4
Vec(X), 91
∆, 31
gra R, 87
ÔX,x , 91
L∞,t0 (f), 140
L∞ (f), 143
∇f, 30
ν(A), 144
157
158 LIST OF NOTATIONS
Note, that some of the notations are discussed in “Suggested notations

for Groups of Polynomial Automorphisms” on page v.
Index
Abhyankar-Moh-Suzuki theorem, 114 embedding, rectifiable, 114

affine parametric line, 148 embeddings, equivalent, 113
affine variety, 51 equidimensional, 78
algebraic vector field on X, 91 equivalent embeddings, 113
associated graded algebras, 87 evaluation, 45
evaluation map, 45
Basissatz, 56 exponential map, 29
bifurcation set, 138
blow-up along f, 133 Fedoryuk’s condition, 140
fiber, 64
canonical symplectic form, 40 fiber, generic, 117
chain rule, 1 field generators, 121
closed subvariety, 52 field of rational functions, 58
Cohn matrix, 6 finite morphism, 71
cone, 46 finite over, 71
constructible set, 80 fixator, 6
contravariance, 1
coordinate polynomial, 150 Gaffney number, 145
critical value, 138 generic fiber, 117
critical value at infinity, 138 generically rational polynomials, 121
generically vs. generally, 117
decomposition theorem, 77 global Milnor fibration, 138
degree conjecture, 26 graded algebra, 87
degree of a morphism, 80 gradient, 30
derivation, 29
derivation, in a point x, 81 hessian, 31
derivation, locally nilpotent, 29 homogeneous, 47
differential of a morphism, 95 homogeneous coordinate ring, 128
discrete valuation of a field, 105 hyperplane at infinity, 128
discrete valuation ring, 106 hypersurface, 49
discriminant, 44
divisor, 111 ideal, 47
divisor class group, 111 ideal of zero set, 47
divisor, principal, 111 integral over A, 101
dominant morphism, 65 intersection multiplicity, 131
irreducible, 55
elementary automorphism, 7 irreducible component, 56
159
160 INDEX
irreducible decomposition, 56 regular, 43, 47

regular function, 43, 47
Jelonek conjecture, 149 regular value at infinity, 138
Jelonek set, 148 residually rational, 124, 131
Jonquière, 7
Jung, van der Kulk, 10 section, 91
Serre criterion for normality, 110
Krull’s principal ideal theorem, 74 smooth variety, 85
special open set, 54
laplace operator, 31 stability, 3
Lin-Zaidenberg theorem, 115 strict transform, 133
local dimension, 69 symmetric product, 52
local homomorphism, 65
local ring, 59 tame automorphism, 8
localization, 59 tangent bundle, 91
locally closed, 79 tangent preserving automorphism, 5
locus at infinity, 128 tangent spaces of fibers, 97
Lojasiewicz number at infinity, 140 tangent vector, 81
Thom theorem, 138
Malgrange’s condition, 140 triangular automorphism, 8
monic distinguished polynomial, 125
morphism of finite degree, 67, 80 uniform boundness statement, 27
morphism, differential of, 95 uniruled varieties, 150
morphism, finite, 71
morphism, of maximal rank, 98 vector field on X, 91
morphism, open morphism, 68 Weierstrass form, 125
nilpotent cone, 46 Weyl algebra, 33
Noether’s formula, 133 Zariski tangent space, 81
Noetherian, 56 Zariski topology, 46, 51
nonsingular variety, 85 zero set, 45, 51
normal variety, 102
Nullstellensatz, 48
one place curve, 123

open morphism, 68
overring principle, 5
plane algebroid curve, 131

Poisson algebra, 35
quadratic transformations, 133
Rabier number, 144

radical, 48
rational function, 58
rational variety, 58
rectifiable embedding, 114

Polynomial Automorphisms and Related Top PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Polynomial Automorphisms and Related Top PDF

Uploaded by

Copyright:

Available Formats

Arno van den Essen

PUBLISHING HOUSE FOR SCIENCE AND TECHNOLOGY

Polynomial Automorphisms and Related Topics

2000 Mathematics Subject Classification: 14E xx, 14 R xx.

Responsibility for Publication: Nguyen Khoa Son

The International School and Workshop “Polynomial Automorphisms and

Ann Arbor, October, 2007

International School and Workshop

Abdus Salam International Centre for Theoretical Physics (ICTP),

Arno Van den Essen, Radboud University of Nijmegen, Netherlands.

Hyman Bass, Michigan University, USA.

Arno Van den Essen, Perhaps the Jacobian Conjecture is simple.

Hanspeter Kraft, An introduction to algebraic transformation.

Shreeram Abhyankar, Some thoughts on the Jacobian conjecture.

Le Tuan Hoa, Finiteness of Hilbert functions and bounds for Castelnuovo-

Jie-Tai Yu, Automorphisms and coordinates of polynomial and free

Wenhua Zhao, Noncommutative symmetric functions.

The International School “Polynomial Automorphisms and Related Topics”

The International Workshop “Polynomial Automorphisms and Related Topics”

for polynomial automorphisms F = (F1, . . . , Fn ) and G = (G1 , . . . , Gn ), we

The general automorphism group. In the spirit of linear algebra, we pro-

The origin preserving group. The group of “origin preserving” automor-

The tangent preserving groups. The group of “tangent preserving” au-

The elementary group. An automorphism is called elementary or de Jon-

The tame group. The group of tame automorphisms, denoted by Tn (R),

Fixator and stabilizer groups. For s = 1, . . . , n, the group GAn−s (R[s] ) is

The stable automorphism group. Finally, we have natural “stability”

Polynomial automorphism groups

1. What is a polynomial automorphism?

3. The case n = 1, R a domain

so u, v ∈ R∗ , v = u−1 , b = −va. Any F = uX + a with u ∈ R∗ is invertible.

which is a homomorphism since

Note that if F ∈ GAn and b = (F1 (0), . . . , Fn (0)), then (X − b) ◦ F =

GAn = Trn · GAon = GAon · Trn

However, this is not a semidirect product. In fact, neither subgroup is normal.

GAon = GLn nGA0n

It follows that GAn = Afn · GA0n (not semidirect).

Proposition 1.3. F ∈ GAn ⇐⇒ G is a polynomial map.

Overring Principle. Suppose R is a subring of R0 , F ∈ MAn (R) with

Proof. Consider the formal inverse.

Wi,n (R) = Afi (R) n GAn−i (R[i] ) .

Conversely (exercise), if F = (a0 + a1X + · · · + ad X d ) with a2, . . . , ad ∈

10. Elementary and tame automorphisms

Theorem 1.4 (Connell). [GA∞ , GA∞ ] is the normal closure of EA∞ .

• Tame automorphisms: The subgroup of tame automorphisms is defined

The proof is an easy exercise.

11. Stable tameness

Figure 1. The triangle ∆ in the XY -plane

12. The case n = 2, R a field

If mn ≤ 1 we have F is linear and in this case it is easy to see that (F, G)

Figure 2. The triangle ∆ in the U V -plane.

with qi (U ) ∈ K = k(U ) and qd (U ) 6= 0.

with µj ∈ k and µe 6= 0, and we define kQk to be µe U e V d . One easily verifies

Step 2 There exist W, Z ∈ hU, V i such that

and any B will work.

Letting c = degV F, d = degV G∗ , then |F|d , |G∗|c are algebraically de-

We must have Y n ∈ supp F , as it is the only point in ∆ having Y -degree

13. Structure of GA2 (k), k a field

For (2), note that (S, T ) = C1 ◦ (P, Q) hence

14. Structure of T2 (R), R a domain

Theorem 1.13. For k a field of characteristic zero, T3 (k) is properly

Theorem 1.14. Considering the containment of GA2 (k [1] ) ⊆ GA3(k) by

Perhaps the Jacobian Conjecture is simple

Example 3.14. The group O2 := {A ∈ M2 | AAt = E} has two irre-